Authors:
- Francis Odum is the Founder/CEO of the Software Analyst Cyber Research where he leads the firm’s research and engagement with cybersecurity leaders.
- Rafal Kitab is a SOC and Incident Response leader at ConnectWise with extensive experience working as a Security Analyst, Engineer, Architect, Incident Responder and recently a Director. He brings considerable experience in Security Operations and frequently shares his takes on his linkedin profile.
Introduction
Cybersecurity operations centers (SOCs) stand at a critical crossroads in the rapidly evolving digital landscape of 2025. This priority emerged as the leading priority amongst CISOs in our network for their organizations.
Security teams are dealing with more alerts than ever before, while also struggling with a shortage of skilled staff. With enterprises now facing thousands of alerts daily, AI has finally transitioned from a stage where it is experimental to now being operational within modern SOC. AI SOC platforms help to monitor systems, investigate alerts, and respond to threats quickly with less manual effort. These tools are helping teams cut down the time it takes to detect and respond to incidents, while also reducing alert fatigue.
With this backdrop, we decided to conduct the most extensive report on this industry. Is this the moment when cybersecurity defenders gain a lasting upper hand, or is the integration of AI simply another fleeting trend in security operations? The insights, analysis, and frameworks presented in this report aim to equip CISOs, security analysts, and organizational decision-makers with the tools to discern hype from substance and make informed strategic choices for the future of their cybersecurity posture.
This report provides an in-depth exploration of the modern SOC ecosystem, specifically analyzing how AI and agentic solutions are transforming the operational capabilities of security teams. We partnered with 13 prominent AI SOC vendors, examining their approaches to automation, alert triage, investigation, and response through rigorous benchmarking and real-world evaluations. We sent each of them a detailed questionnaire and asked them to explain how their platform works. We then used their answers to compare and map their capabilities using our own assessment framework. This report includes summaries of each vendor and highlights the differences in their approaches, features, and overall strengths.
There are many other competitors on the market we have not evaluated, but plan to do so in a future episode. Some vendors that were not evaluated include Tines, SevenAI, Elastic AI, AIStrike, Bricklayer, Simbain and many more building agentic solutions for the SOC.
Market Players (Overview)
The following AI SOC vendors participated in our research and are listed in alphabetical order.
Command Zero is a cybersecurity platform designed to accelerate and automate security investigations through natural language interaction and structured reasoning. The platform integrates a retrieval-augmented generation (RAG) architecture with security-specific reasoning tools, enabling analysts to perform end-to-end investigations using plain English prompts. Rather than focusing on alert triage or basic filtering, Command Zero is purpose-built for handling Tier 2 and above investigations. Its core use case is helping analysts understand the full scope, context, and impact of complex incidents that go beyond initial detection. The platform operates as a reasoning layer that connects to existing security and IT infrastructure. It integrates with a wide range of enterprise systems, including SIEMs, EDRs, identity providers, ticketing tools, and asset inventories. Analysts initiate investigations by describing a question or incident in natural language. The system then retrieves and synthesizes relevant data from these connected tools. This enables investigation-level workflows such as tracing lateral movement, verifying persistence mechanisms, or confirming multi-stage compromise, without requiring manual queries across fragmented interfaces. Command Zero’s investigation engine combines three core functions: retrieval, reasoning, and automation. It retrieves relevant information from underlying systems, applies structured logic to interpret that data in a security context, and automates investigative workflows. The system produces both natural language summaries and structured outputs that can be shared or escalated. This approach supports the depth and rigor required for high-fidelity investigations that would typically be performed by experienced analysts. A key feature of the platform is its ability to generate complete investigation reports as a byproduct of the investigative process. These reports contain all relevant evidence, reasoning, and findings in a format suitable for documentation, collaboration, and escalation. By automating report generation, the platform reduces the time analysts spend compiling results and ensures consistency in investigative outputs across the team. Command Zero does not serve as a detection or triage system. It is designed to activate once an alert has been generated or a security question has been raised. This distinction makes it suitable for Tier 2 and Tier 3 analysts who are tasked with answering what happened, how it happened, and what the potential impact is. The platform complements existing alerting systems by offering investigation depth rather than alert suppression or prioritization. To support analysts across varying levels of experience, Command Zero abstracts away query languages and tool-specific logic. Analysts interact with the system using plain English, while the platform handles the complexity of retrieving and correlating data from diverse sources. This helps less experienced analysts operate at a higher level of sophistication while enabling senior analysts to scale their expertise across more cases. Deployment options include customer-hosted and cloud-hosted configurations. Command Zero integrates via APIs with existing infrastructure, querying data in place rather than ingesting it. This minimizes duplication, reduces operational overhead, and ensures compatibility with existing security investments. Customization is enabled through environment-specific connectors and structured action libraries. Organizations can define how the platform interacts with their systems and tailor its behavior to match internal workflows and terminology. This allows teams to align the platform with their investigation procedures and response protocols. In summary, Command Zero provides a retrieval-augmented investigation platform purpose-built for Tier 2 and Tier 3 analysts. Its focus on natural language reasoning, automated report generation, and cross-system data retrieval enables security teams to accelerate complex investigations while maintaining consistency, depth, and operational efficiency. Crogl is a cybersecurity company that provides a software platform focused on automating alert triage and incident management within enterprise Security Operations Centers (SOCs). It positions itself as an alternative to both managed detection and response (MDR) services and traditional Security Information and Event Management (SIEM) platforms. The system is built to streamline detection engineering, investigation workflows, and analyst productivity through a fully integrated artificial intelligence (AI) and data stack. The foundation of Crogl’s platform is a compound AI system that includes a knowledge engine that automatically learns the customer’s data environment, without requiring data normalization or movement. It connects to all security relevant data sets and learns their schemas, including security telemetry, cloud and identity logs, and human analyst input such as process documentation or ticket annotation. It includes threat intelligence from security advisories. Crogl does not rely on rule-based correlation or static detection signatures. Instead, it uses a series of AI agents that continuously process security events and classify activity in real time. These agents operate on normalized timelines that include events from multiple sources, providing context for accurate threat assessment and decision-making. A key aspect of Crogl’s system is its learning-based approach to triage. When new alerts are received, the platform attempts to classify them based on prior data and learned patterns. It autogenerates response plans based on industry best practices, and learnings from prior analyst processes, and executes them surfacing evidence and documentation at each step. If confidence is high, the alert is suppressed or resolved automatically. If not, the system surfaces the alert to an analyst as an investigation. Analyst decisions, including annotations and feedback, are captured and used to refine future detection and response behavior. This forms a continuous learning loop where the system evolves based on local operational knowledge. Crogl emphasizes the generation of structured case timelines that consolidate related events. These timelines integrate metadata such as user and asset context, behavioral indicators, and external threat signals. Analysts can interact with these timelines directly within the platform, conduct investigations, and document findings without switching tools. The interface is built for collaborative workflows, allowing for shared views and contextual notes among team members. The platform is designed to integrate with existing SOC tooling and processes. It offers APIs for alert forwarding, case ticketing, and automated remediation. Out-of-the-box integrations with common platforms like XSOAR, Splunk, Jira, Crowdstrike and allow security teams to embed Crogl’s capabilities into their day-to-day operations without major changes in workflow. Crogl is designed to reduce the volume of alerts requiring human attention. According to reported performance in customer environments, more than 95 percent of incoming alerts are either suppressed or resolved automatically. This reduction is achieved through AI-based contextual analysis rather than static filtering. The platform uses environmental understanding and behavioral context to decide which events need escalation, preserving fidelity while reducing noise. From an infrastructure standpoint, Crogl supports flexible deployment options. Customers may choose to use their own cloud object storage, default to Crogl’s managed infrastructure, or deploy the platform fully on-premises in air-gapped environments with no Internet access. This flexibility allows organizations to meet specific compliance, sovereignty, or operational requirements. The pricing model is based on the number of monitored workspaces or log sources rather than data volume, making the platform scalable without variable cost penalties based on ingestion rates. Customization is driven by analyst interaction and feedback. Unlike platforms that require manual tuning or rule creation, Crogl uses customer input and best practices as training data for its models. However, organizations can still apply custom business logic, exclusions, or policy constraints to guide alert disposition and suppression. In summary, Crogl delivers an AI-based SOC automation platform that focuses on intelligent alert triage, integrated investigation support, and analyst-driven learning. Its emphasis on adaptive response, structured data correlation, and operational integration provides a framework for modernizing SOC functions without relying on external service providers or traditional rules-based systems. D3 Security is a cybersecurity automation vendor focused on enabling AI-driven operations within security teams. Its platform, Morpheus, is an Autonomous SOC solution that combines automation, adaptive logic, and cross-stack correlation. It is designed for both enterprise SOCs and managed security service providers. Dropzone is a cybersecurity company that provides an AI-powered platform focused on reducing alert fatigue and improving incident response efficiency within Security Operations Centers (SOCs). Its primary function is to automate the triage of alerts generated by existing security tools, while maintaining high fidelity through a structured quality assurance process. Dropzone positions itself as a complement to established detection infrastructure, offering integration with a wide range of alert sources such as EDR platforms, SIEMs, cloud logs, and identity providers. The platform is built around an AI system trained to mimic and scale human analyst behavior. Each alert received by the system is enriched and investigated through a series of structured reasoning steps, with results summarized in a natural language format. These summaries provide evidence-based justifications for classification decisions, helping analysts quickly assess whether an alert requires escalation. Alert dispositions are categorized along a confidence spectrum, typically ranging from benign to confirmed malicious, with appropriate context included for each outcome. A distinguishing feature of Dropzone’s approach is its emphasis on human-in-the-loop quality control. Unlike black-box AI systems or fully automated triage layers, Dropzone designs its user interface to make it easy for human analysts to review investigations and provide feedback, if needed. Dropzone incorporates a dedicated QA process that systematically samples and reviews the output of its AI models. This QA function is performed by a team of experienced analysts who validate alert decisions, assess correctness, and provide structured feedback Dropzone integrates directly into customer environments via APIs and does not require changes to existing detection rules or logging pipelines. It consumes alerts from platforms such as CrowdStrike, SentinelOne, Microsoft Defender, and Okta, among others. The system acts as a triage layer between raw alerts and case management platforms like Jira, Splunk, or ServiceNow. Customers retain control over how Dropzone’s decisions are operationalized, with options to automatically close low-confidence alerts or escalate high-confidence alerts directly into ticketing systems. Deployment options include cloud-hosted and hybrid models. Customers can bring their own storage or use Dropzone’s managed infrastructure. The platform is designed to be lightweight to deploy, with initial onboarding often completed in under a week. Pricing is typically based on the volume of alerts triaged, rather than log ingestion or storage capacity. Customization is available through environment-specific tuning and policy controls. Customers can define business logic for alert disposition, suppression, or routing. Additionally, the platform supports tagging and feedback mechanisms, which analysts can use to correct or reinforce model behavior. These inputs are incorporated into system updates via the QA loop. The QA process operates on a near-daily cycle, with performance metrics such as false positive rates, investigation completeness, and reasoning quality tracked over time. Dropzone uses this data to tune both its prompts and the workflows that guide automated investigations. Dropzone stores details about the customer environment and business learned during investigations, providing RAG context to improve future investigations. Customers can also add details to this context memory database and report this feature improving accuracy. In summary, Dropzone offers an AI-powered alert triage platform focused on high-quality, analyst-validated decisions. Its structured QA process, combined with real-time alert reasoning and transparent summarization, allows organizations to reduce manual workload without compromising accuracy. By prioritizing model oversight and operational trust, Dropzone provides a rigorous framework for integrating AI into SOC workflows with measurable performance assurance. Exaforce is an agentic AI SOC platform designed to support the full security operations Exaforce also augments the investigation interface with its use of investigative graphs. Intezer is a cybersecurity platform that provides end-to-end support for threat detection, investigation, and response, positioning itself as a complete AI SOC system. While the platform includes broad capabilities for automated alert triage, threat enrichment, and case management, its standout strength lies in its advanced sandboxing and deep malware analysis engine. This foundation enables Intezer to offer high-confidence investigations, particularly in environments dealing with malware-heavy threats or nation-state level adversaries. At the core of the platform is a dynamic sandbox integrated into the broader SOC workflow. This sandbox executes suspicious files and payloads in a controlled environment and captures detailed behavioral indicators, including process creation, memory injection, persistence techniques, and network communication. Rather than simply flagging based on signatures or basic heuristics, the sandbox enables high-fidelity behavioral analysis that directly informs automated decisions and escalations. The sandbox is tightly integrated with static analysis, code reuse detection, and memory forensics, allowing for a comprehensive understanding of threats at multiple levels. Intezer’s investigative approach is informed by years of experience in reverse engineering and malware classification. The platform’s core engine dissects binaries into code fragments and compares them to a proprietary database of known code from legitimate software, malware families, and threat actor toolkits. This allows the system to identify reused components across malware campaigns, even in polymorphic or obfuscated variants. As a result, Intezer can make high-confidence verdicts on new and unknown threats by tracing their code origins and matching behavior against known profiles. Intezer functions as a full SOC automation platform. It ingests alerts from existing tools, including EDRs, SIEMs, and email security platforms, and applies layered analysis to surface only high-fidelity incidents. The system automatically enriches alerts with sandbox results, memory scan indicators, MITRE ATT&CK techniques, and threat intelligence. Analysts receive structured investigation reports that combine behavioral summaries, classification context, and recommended actions, allowing them to understand and respond to incidents without pivoting across tools. Case management is built into the platform, enabling teams to track investigations, assign tasks, and collaborate within the system. Analysts can submit files, memory dumps, or suspicious processes manually or through automated workflows. Each investigation benefits from the same deep analysis pipeline, whether triggered by an endpoint alert or an email attachment. Verdicts are recorded and shared across the team, improving consistency and organizational memory over time. The platform offers broad integrations across Cloud, Email, Endpoint, SIEM, SOAR, Identity, Network and Ticketing environments supporting both fully automated and analyst-in-the-loop workflows. Customization is supported through tagging policies, trusted software lists, suppression rules, and automation playbooks. Deployment options include cloud-hosted and private cloud configurations, enabling scalability for enterprise and MSSP use cases. While Intezer provides broad SOC coverage, its sandbox and malware analysis engine remains a central differentiator. This capability allows the platform to deliver depth on threats that would otherwise evade lightweight analysis, including fileless malware, in-memory implants, and customized payloads. For organizations facing persistent or sophisticated threat actors, Intezer offers an AI SOC platform with a malware investigation core that maintains the fidelity and investigative rigor often missing from generalized detection tools. In summary, Intezer is a full AI SOC platform that integrates alert ingestion, analysis, and response, with a particular strength in sandbox-driven threat understanding. Its combination of behavioral analysis, code origin tracing, and memory forensics delivers comprehensive coverage for complex attack scenarios, making it a strong option for security teams prioritizing depth and accuracy in malware investigations. Legion is a cybersecurity platform that enables security teams to codify and automate investigative workflows by capturing how analysts work and transforming those behaviors into reusable agents. Rather than relying on predefined detection rules or generalized models, Legion focuses on recording real analyst decision-making processes and operationalizing them at scale. At the core of Legion’s platform is a browser-based agent that records analyst sessions. As analysts investigate alerts, the system captures the full investigative process, including data reviewed, steps taken, and decisions made. This recording is not just for playback or training purposes. Instead, it becomes a living agent that can be reused in future investigations. This design allows teams to apply their best investigative logic to every new alert, effectively scaling the expertise of senior analysts across the entire SOC. These recorded agents can be tested, refined, and re-executed. They are not static or one-time scripts but instead act as repeatable workflows that maintain context across investigations. Legion supports investigation reuse by allowing agents to automatically execute on new, relevant alerts—applying the same reasoning process that a human analyst previously followed. As a result, teams can gradually build a library of agents that reflect their institutional knowledge and investigative standards. Legion integrates into security environments by operating as a system of record for investigations. It pulls in alerts from existing detection systems and provides a workspace where analysts can review, annotate, and act on these alerts. Over time, the system learns which investigation patterns lead to high-confidence outcomes and encourages reuse of those patterns. Deployment flexibility is supported through cloud-hosted, customer-hosted, and hybrid options. This allows organizations with different infrastructure and compliance requirements to adopt the platform without sacrificing control or performance. Customization is inherent to Legion’s design. Every agent begins as an analyst-driven investigation, meaning each workflow is tailored to the specific environment, threat landscape, and operational style of the customer. Rather than generalizing across organizations, Legion’s model emphasizes internal accuracy and context-specific decision-making. In summary, Legion provides a framework for transforming analyst expertise into operational automation. By recording and replaying investigations through browser-based agents, it enables security teams to standardize, scale, and continuously improve how alerts are handled. The platform focuses on investigative quality, contextual fidelity, and repeatability, offering a unique approach to SOC automation rooted in real analyst behavior. Mate is a security automation platform designed to scale and operationalize the investigative expertise of human analysts. While often described as a “record and replay” system, this characterization significantly understates its architectural depth. While browser-based recording remains a valuable source of observational data, it is treated as one input among many. Recorded workflows include data lookups, cross-tool navigations, tagging logic, and response actions, all contextualized by correlated inputs from structured backend systems. These workflows are then reviewed, edited, and versioned before being deployed, ensuring traceability and oversight. As analysts work through new alerts, the platform continuously expands its automation library and adapts based on ongoing feedback. Mate is capable of resolving routine alerts autonomously and accelerating more complex investigations by surfacing rich contextual data and proposing recommended actions. Because of its multi-source architecture, Mate is not limited by tool-specific integrations. It works across third-party platforms and organizational systems, even in environments where formal APIs are not present. This independence from integration constraints enables rapid onboarding and broad applicability across environments with heterogeneous tooling. The platform also supports user feedback as a live input into its learning model. Analysts can accept, reject, or modify automated actions, and Mate incorporates this feedback automatically into its reasoning. Over time, this creates a feedback loop where human oversight drives continuous refinement without the need for manual rule tuning or retraining. Additionally, business context, such as frequent traveler status, role-specific activity, or privileged account usage, can be factored into decision-making, helping to reduce false positives and better reflect organizational norms. Deployment can be cloud-hosted or hybrid, with all automation and data handling confined to the customer’s controlled environment. This ensures data sovereignty while maintaining the flexibility and speed of a modern security automation platform. Mate’s approach delivers value from day one. Unlike conventional automation systems that require weeks of rule development or machine learning models that depend on large training datasets, Mate’s reasoning engine and contextual awareness allow it to begin providing meaningful assistance during the earliest phases of deployment. In summary, Mate is a multi-source, context-aware automation platform that combines behavioral observation with backend integration, reason-based learning, and judgment-driven validation. By focusing on why analysts act, rather than just what they do, and by incorporating feedback and business context continuously, Mate delivers high-fidelity automation that aligns with real-world security operations needs from the outset. Its unique architecture allows organizations to scale analyst expertise without compromising accuracy, control, or adaptability. Prophet Security is a cybersecurity platform (Prophet AI) designed to automate alert triage, investigation, and incident response. It provides an AI-driven system that ingests alerts from existing detection tools and applies automated reasoning to determine their severity, relevance, and context. Built for modern Security Operations Centers (SOCs), Prophet Security aims to reduce the volume of alerts requiring manual analyst review by resolving or escalating them based on high-confidence analysis. The platform’s core focus is on investigation fidelity, combining structured workflows with machine learning models and large language model (LLM) capabilities. Prophet AI ingests alerts from tools such as CrowdStrike, SentinelOne, Okta, and Microsoft Defender as well as custom detections from SIEMs . Once alerts are received, the system conducts contextual enrichment using telemetry from endpoints, cloud infrastructure, identity systems, and threat intelligence. Rather than acting as a rule-based filter, Prophet AI applies automated investigation logic designed to replicate how experienced analysts would approach each case. The platform reviews user behavior, asset roles, recent activity patterns, and known threat indicators to determine whether an alert is benign or requires further investigation. At the heart of Prophet’s approach is the concept of “resolving” alerts through agentic reasoning . The system generates structured, evidence-based reports for each alert, summarizing findings in natural language and citing the data sources used. These investigation reports are designed to be human-readable and ready for audit, documentation, or escalation. The reports include confidence levels, reasoning steps, and relevant context such as attack techniques mapped to the MITRE ATT&CK framework. Prophet AI supports multiple modes of operation, including fully automated resolution, analyst-in-the-loop confirmation, and automatic escalation to case management systems. This flexibility allows organizations to phase in automation based on trust, maturity, or compliance requirements. Security teams can configure policies to automatically close low-risk alerts, forward medium-risk alerts for review, or escalate high-risk findings with full investigative context. Additionally, Prophet AI offers copilot-style capabilities where users can ask their own investigative questions in free form, natural language, both in the context of individual investigations as well as more open ended threat hunts. The platform is built to support the needs of Tier 1 through Tier 3 analysts. Junior analysts benefit from structured reports that provide visibility into investigative reasoning, while senior analysts can rely on automation to reduce the burden of repetitive triage. Prophet’s use of LLMs supports summarization, data correlation, and contextual reasoning, enabling faster understanding without compromising accuracy. Integration with existing SOC tooling is a key part of the deployment model. Prophet AI connects with alerting systems, ticketing platforms, SIEMs, and SOAR tools via API. Alerts can be ingested in real time, with investigations triggered automatically or manually based on use case. Deployment options include cloud-hosted and private infrastructure models, allowing for flexibility across enterprise environments. Customization is supported through tuning policies, alert handling logic, and feedback mechanisms. Analysts can tag misclassified alerts, adjust thresholds, or define business-specific logic for disposition. Prophet incorporates these inputs to refine its reasoning over time, improving performance in dynamic environments. In summary, Prophet Security offers an AI-driven SOC platform focused on resolving alerts through automated investigation. Its strength lies in delivering structured, auditable findings for each alert while reducing the time and effort required from analysts. With support for real-time enrichment, LLM-powered reasoning, and flexible automation workflows, Prophet provides a scalable solution for organizations looking to improve SOC efficiency without sacrificing investigative quality. Qevlar is an AI-driven security platform built to assist analysts in investigating alerts with greater speed, consistency, and accuracy. Rather than functioning as a full alert triage system or replacing detection tools, Qevlar is positioned as an investigation co-pilot that enhances the analyst workflow by automating the reasoning and documentation processes. The platform integrates into existing SOC environments and is designed to support analysts in making informed decisions while reducing the time and effort spent on repetitive tasks. Qevlar operates by ingesting alerts from various sources such as EDR platforms and identity providers. Upon receiving an alert, the platform performs an automated investigation that mimics the type of reasoning and research an analyst would typically conduct. It produces a structured report for each alert, outlining whether it is benign or malicious, providing supporting evidence, and assigning a confidence level to the conclusion. This format is designed to be immediately usable by analysts, team leads, or case management systems and includes links to the data sources used in the decision process. Unlike traditional automation platforms, Qevlar does not aim to suppress or prioritize alerts at the point of ingestion. Instead, it focuses on enriching and interpreting alerts after they have been generated, ensuring that analysts receive clear, context-rich findings for review or action. The system is built to preserve analyst oversight, making it well-suited for SOCs that require human-in-the-loop validation, high-confidence decisions, and audit-ready outputs. A key design feature of Qevlar is its ability to automate the documentation process. Each investigation results in a report that includes a plain-language summary of findings, references to supporting data, and a justification for the final verdict. These reports are structured and consistent, helping teams reduce the manual burden of writing investigation notes or escalation summaries. This also supports faster peer review, incident handoff, and knowledge sharing across the SOC. Qevlar integrates with a range of commonly used security tools. The platform supports ingestion of alerts from sources such as SentinelOne, CrowdStrike, Microsoft Defender, and Okta, and can export investigation results to case management and ticketing systems like Jira or TheHive. This allows organizations to embed Qevlar’s automated reasoning into their existing workflows without significant disruption. Deployment options include cloud-hosted and fully on-premises models. The on-premises option allows Qevlar to operate in environments with strict data residency, sovereignty, or compliance requirements. Regardless of deployment model, investigation data remains within the customer’s infrastructure, and the system does not require log ingestion or large-scale data duplication. Qevlar is designed to evolve with the environment it operates in. The platform supports feedback loops where analysts can correct outcomes, tune decisions, and provide input on edge cases. These corrections are incorporated into future investigations, improving the system’s ability to align with internal policies and expectations. This enables a continuous improvement cycle without needing complex rule writing or manual tuning. In summary, Qevlar is an investigation-focused AI platform that acts as a co-pilot for SOC analysts, providing structured, explainable findings for each alert without removing human control. Its emphasis on automated reasoning, report generation, and easy integration into existing environments makes it a strong fit for teams looking to increase investigative capacity while maintaining accuracy, transparency, and adaptability. With support for on-prem deployment and analyst feedback, Qevlar offers a high-trust automation layer tailored to the needs of modern security operations. Radiant Security, founded in 2021 by former Exabeam executives, is a cybersecurity company focused on automating Security Operations Center (SOC) triage and response through artificial intelligence. Its core offering is an AI-driven SOC platform that seeks to address alert fatigue, reduce manual investigation workloads, and improve response time to security incidents. Radiant’s system is designed to triage any security alert (including SIEM alerts) by integrating with existing enterprise environments and ingest telemetry from multiple sources, including alerts, endpoint logs, identity platforms, and any unstructured security telemetry and documentation. The platform operates via a fleet of AI agents that perform automated triage and investigation across a wide array of security use cases. Its principal mechanism involves contextual enrichment of alerts, where extracted data artifacts are dynamically correlated with environmental signals such as user roles, behavioral baselines, and threat intelligence. Radiant employs just-in-time behavioral baselining rather than precomputed ML models, enabling investigations to adapt to context-specific activity patterns without continuous background processing. These baselines are generated from recent data, typically ranging from two weeks to 90 days, depending on the nature of the event. Radiant distinguishes between known and unknown threats using a tiered model. For novel or previously unseen alerts, Radiant conducts ad hoc analysis and constructs new triage procedures, effectively augmenting the system’s capabilities over time. Once analyzed, alerts are categorized into one of four statuses: benign, recommended benign, recommended malicious, or malicious. Only alerts deemed malicious or confirmed by a user are escalated to case-level incidents, helping reduce operational noise. A significant focus is placed on automation. Radiant supports full-cycle automated remediation workflows, which can be configured to include or exclude human approval steps. This hybrid model allows organizations to gradually shift toward increased automation as trust in the system grows. Radiant’s internal benchmarks suggest that full triage can occur within minutes, though mean time to respond (MTR) is often limited by user validation time. Customers reportedly reduce response cycles from industry-standard durations of several days to under eight hours, with automation-capable environments achieving even faster turnaround of as short as a few minutes. Radiant’s deployment model supports customer-controlled log storage, typically through Amazon S3 buckets as well as query connectors that leverage existing SIEMs and APIs. This configuration allows for compliance-friendly, long-term retention without the cost burden of traditional SIEM licensing models. The platform includes a built-in log management interface with support for Lucene queries and Grafana integration. This approach enables visualization and correlation without proprietary query languages or infrastructure dependencies. Licensing is decoupled from data volume and is instead based on the number of users or alert types under management. This model aims to provide predictable pricing while scaling with organizational complexity and coverage requirements. For managed security service providers (MSSPs), Radiant offers licensing aggregation across tenants, with multi-bucket support to preserve data isolation and ownership. From a customization standpoint, Radiant offers three primary levers: static allow/deny lists, environment-specific unsupervised learning, and codified natural language policies. These options allow organizations to tailor alert disposition logic to their operational context and reduce false positives, particularly in environments where technical teams may generate anomalous but legitimate activity. Radiant positions itself as a product-centric platform rather than a managed service. Customers are responsible for deployment and operation, with support limited to technical guidance. Initial implementation typically includes a four to six-week proof of concept, followed by a three-month “time to trust” period during which automation is incrementally enabled. In summary, Radiant Security delivers an AI-based SOC automation platform focused on reducing alert volume, expediting investigation, and supporting scalable automation. Its emphasis on adaptive triage, just-in-time baselining, and customer-owned data infrastructure offers a distinct model within the AI SOC landscape. Sekoia.io is a European cybersecurity vendor that delivers a unified SOC platform integrating SIEM, SOAR, XDR, and Cyber Threat Intelligence (CTI) capabilities under a single architecture. Founded in France and now expanding globally, the company has raised over €60 million in funding to date, signaling strong investor confidence in its product vision and operational execution. The platform’s focus is on reducing the operational burden on SOC teams by combining structured data ingestion, automated detection, and AI-enhanced workflows, all with native threat intelligence at its core. Sekoia’s primary offering is its AI-powered SOC platform, Sekoia Defend, which is tightly integrated with Sekoia Intelligence, a standalone threat intelligence module that can also be used independently. Together, these components provide flexible deployment options and align well with the needs of both enterprise SOCs and MSSPs. Sekoia’s MSSP strategy, in particular, is a major pillar of its go-to-market approach, supported by deployments across Europe. A defining feature of the platform is its approach to AI integration. Rather than focusing on full automation or replacing analysts, Sekoia’s AI is designed to work alongside human operators. AI features assist with alert enrichment, contextual grouping, tuning recommendations, and providing intelligent summaries, aiming to reduce alert fatigue while retaining human oversight. One example of this is “Roy,” Sekoia’s AI chatbot, which enables natural language interactions across use cases ranging from threat intelligence lookups to alert triage and detection logic. Roy supports threat hunting, assists with query generation, and can explain threat actor campaigns, helping analysts work faster without needing deep familiarity with the underlying tooling. Sekoia’s detection capabilities span multiple layers. These include: All detection outputs are mapped to the MITRE ATT&CK framework, providing structured visibility into how threats manifest and enabling SOC teams to assess and prioritize coverage gaps. The platform is designed to ingest telemetry from a wide range of sources, including cloud providers, endpoint tools, and network sensors. Its open-standards approach ensures compatibility with other tools in the security ecosystem: CTI is stored as STIX objects, and network telemetry follows the OCSF format. This data-first design simplifies integration and supports more precise correlation and enrichment during investigations. Sekoia also provides a query-based threat hunting interface, supporting syntax similar to Microsoft’s KQL. Analysts can craft complex searches or rely on Roy to assist with hunt logic and execution, allowing both junior and senior team members to contribute effectively to proactive threat identification. On the response side, Sekoia currently supports playbooks and SOAR-style automation, with AI-driven remediation capabilities under development. While the current system avoids fully automated alert closure, upcoming releases will introduce customizable AI agents that support guided remediation actions, designed with varying degrees of autonomy to accommodate organizational risk tolerance. From a commercial standpoint, Sekoia offers both per-asset and volume-based pricing models. The per-asset model is favored by customers for its predictability and affordability relative to traditional SIEM licensing. Additionally, Sekoia is optimizing its storage backend to reduce long-term retention costs, enabling customers to retain more searchable data beyond the default 90-day window. In summary, Sekoia.io offers a full-spectrum AI SOC platform that combines detection, investigation, response, and threat intelligence into a tightly integrated system. Its analyst-centric AI design, native CTI engine, and strong support for open standards position it as a flexible and scalable solution for modern SOC environments, particularly those prioritizing speed, transparency, and contextual depth in security operations. Torq is a security automation platform that enables organizations to build, manage, and scale workflows across the security operations lifecycle without writing code. Designed for SOC teams, incident responders, and security engineers, the platform provides an orchestration layer that connects tools, data sources, and decision logic into real-time automated processes. Its core value lies in accelerating detection-to-response cycles, reducing manual workload, and enabling consistent operational execution through a modular and extensible design. The platform’s foundation is a no-code visual editor that allows users to build automation workflows by assembling actions, triggers, and decision points. These workflows, referred to as “hyperautomation pipelines,” can span across SIEMs, EDRs, identity providers, cloud platforms, ticketing systems, and threat intelligence feeds. Security teams use these pipelines to automate tasks such as alert triage, data enrichment, case assignment, threat containment, and remediation. The system supports real-time execution of workflows, enabling immediate response to new alerts or indicators. Torq includes over 1,000 prebuilt integrations with widely used security and IT tools. These include platforms such as CrowdStrike, SentinelOne, Microsoft Defender, Okta, ServiceNow, Jira, and Slack. The integration framework is API-first and extensible, allowing teams to add new tools or custom connectors without disrupting existing workflows. This makes the platform adaptable across different environments, including enterprises with diverse tooling or MSSPs managing multiple tenants. The platform is designed to support dynamic, event-driven automation. Triggers can be based on incoming alerts, scheduled tasks, analyst inputs, or external API calls. Workflows can incorporate conditional logic, parallel branches, and data transformations, allowing for complex decision trees and branching behavior. Torq’s architecture enables workflows to evolve over time, with version control and audit logging to track changes and maintain compliance. One of Torq’s key design principles is usability for both technical and non-technical users. Security engineers can build and maintain workflows without needing to write scripts, while analysts can interact with them through forms, approvals, and notifications. This enables a collaborative operating model where different SOC roles can contribute to and benefit from automation without siloed ownership. Torq also supports templated “use cases” that bundle workflows for common security tasks. Examples include phishing response, malware containment, privilege escalation review, and suspicious login analysis. These use cases can be deployed and customized quickly, accelerating time to value for teams seeking to operationalize automation without starting from scratch. Deployment options include cloud-hosted and customer-hosted environments, with features for multi-tenant isolation and enterprise access controls. The platform is built to scale across large volumes of alerts, users, and integrations, with monitoring and performance dashboards to support operational management. Customization is central to Torq’s value proposition. Every workflow can be adapted to match an organization’s environment, policies, and processes. Data inputs and outputs can be normalized, enriched, or transformed as needed, and analyst inputs can be integrated at any stage of execution. This flexibility allows organizations to maintain high trust in automation outcomes while preserving the ability to intervene or escalate when required. In summary, Torq provides a scalable security automation platform that allows SOC teams to build and run real-time workflows across their tool ecosystem. Its no-code design, extensive integrations, and support for dynamic decision-making make it well suited for organizations looking to operationalize automation without the complexity of traditional scripting or orchestration tools.
Command Zero
2022
Crogl
2023
D3 Security
2025
The core functionality of Morpheus is its dynamic, AI-driven playbooks, which can adjust their behavior during investigations based on contextual changes. These playbooks incorporate both horizontal correlation across different tools, such as SIEM, EDR, IAM, and cloud platforms, and vertical analysis within time-series or source-specific patterns. This enables the platform to investigate and triage threats at machine speed with minimal human intervention, improving response consistency and reducing analyst workload.
Morpheus supports over 700 integrations across endpoint, cloud, identity, ticketing, and threat intelligence platforms. These integrations are maintained and versioned through a modular framework that ensures long-term stability and makes it easier to adapt to evolving tech stacks. This integration depth allows security teams to operationalize automation across diverse environments and rapidly onboard new tools without extensive engineering effort.
A key feature of Morpheus is its data normalization and correlation layer. Alerts from different systems are standardized into a common format, enabling unified analysis, de-duplication, and automated enrichment. This makes it easier to group related alerts, draw faster conclusions, and leverage asset data, vulnerability intelligence, and historical incident records.
Morpheus includes a built-in chatbot that allows users to query, orchestrate, and investigate across the stack using natural language. Analysts can interact with the platform in plain English, eliminating the need for coding or familiarity with query languages like KQL. This capability shortens investigation times and makes key functions accessible to junior analysts and non-technical users.
While Morpheus is designed for autonomous operation, it also incorporates functionality developed during D3’s history as a SOAR vendor. These include robust case management, role-based access control, SLA tracking, audit trails, and executive dashboards. Each case can be linked to the automation that triggered it, ensuring traceability from alert to resolution and supporting compliance and oversight requirements.
The platform offers flexibility in deployment, with support for cloud-hosted, hybrid, and on-premises models. Multi-tenant capabilities allow MSSPs to maintain customer separation while centrally managing playbooks and response logic. Customization is supported at multiple layers, including playbook logic, UI configuration, and data handling, allowing teams to align automation with internal policies and update workflows based on real-world experience.
Morpheus actively performs investigations and triage, and supports remediation, across the security stack. By correlating alerts, executing automated responses, and enriching context on the fly, it delivers faster and more accurate outcomes than manual processes. The platform significantly reduces analyst fatigue by resolving routine alerts automatically and also relieves engineering teams of scripting and maintenance overhead. This makes security operations more scalable, more consistent, and ultimately more effective in preventing and containing threats.
Dropzone AI
2023
Exaforce
2023
lifecycle security operations. Its core functionality centers on AI Agents, called Exabots,
and AI-powered data exploration capabilities that enable organizations to reduce alert
overload, speed up investigations, expand detection coverage and reduce spend across
triage, detection, investigation and response.
The defining aspect of Exaforce’s approach is a multi-model AI engine, purpose-built for
cybersecurity use cases where massive volumes of real-time data must be continuously
analyzed. The multi-model AI engine combines three types of AI/ML models. It ingests
and contextualizes data without the need for a SIEM, which reduces the volume of data
sent to SIEMs, cutting storage and licensing costs. It uses machine learning to detect
anomalies and reduce false positives through behavioral context, and applies large
language models for analysis and task execution.
Unlike platforms that require extensive configuration or rule tuning to provide value,
Exaforce’s agents are pre-trained and are designed to deliver significant out-of-the-box
effectiveness. Deployment options include hosted, self-hosted, or MDR models, giving
organizations flexibility based on infrastructure requirements or data governance
constraints. Upon deployment, the system is capable of triaging a wide range of alert
types, applying baseline detection and enrichment logic to provide high confidence
recommendations. This allows organizations to accelerate time-to-value and begin
reducing manual analyst workload early in the adoption process.
SOC teams interact with Exaforce via a web-based interface that presents findings,
timelines, alert context, and asset relationships in a single view. Analysts can review the
sequence of events leading to an alert, validate automated findings, and initiate
response actions directly from the interface. Analysts can take action, escalate, or
annotate directly within the interface. Notably, user feedback on alert quality is
automatically incorporated into the system’s decision-making AI models during future
analysis. This adaptive feedback loop helps further reduce false positives over time and
ensures that the platform remains aligned with operational needs as environments
evolve.
In addition to feedback, Exaforce allows for customization via business context via
natural language into its models. For example, user roles, behavioral baselines, or
attributes such as frequent travel status can be included in decision-making to minimize
noise from legitimate activity. This ability to incorporate contextual awareness into
detection and triage processes helps improve accuracy and reduces unnecessary
escalations. The system’s AI models allow for continuous tuning and new telemetry
sources can be onboarded with minimal friction. This makes the platform adaptable to
changing environments, organization specific details, and evolving threat profiles.
These graphs are automatically generated as the system observes related events
across the environment. Each graph connects assets, users, alerts, and behaviors to
surface meaningful relationships and investigative leads for powerful visualizations of
users, assets and events.
Overall, Exaforce offers an AI SOC platform focused on delivering full lifecycle security
operations, across alert triage, investigations, threat detection and response. The
platform can be leveraged by security organizations who do not have mature SOC
tooling (SIEM, SOAR) to accelerate their journey to building an effective SOC practice
or more mature SOC organizations to improve the productivity and efficacy of their SOC
analysts, detection engineers and threat hunters.
Intezer
2017
Legion
2024
Mate
2025
Prophet Security
2023
Qevlar AI
2023
Radiant Security
2021
Sekoia
2020
Torq
2020
To view the complete scoring, head over to this spreadsheet.
The Modern SOC
Data Fabric Layer
In SOC teams, a layer of raw security data from diverse sources must be standardized and processed for effective analysis before being sent to a SIEM (or automation solution). This category has two distinct types of vendors. 1) Those focused on building and managing the engineering data pipeline. 2) Those prioritizing data filtering and enrichment to improve detection quality. The key task is formatting data into a consistent structure to enable seamless integration and adding contextual information, such as IP geolocation or threat intelligence feeds, to improve data quality. We’ll uncover how this layer increasingly intersects with automation efforts.
Storage and Detection Layer
Companies leverage threat detection rules to define the logic for identifying malicious activity in their data. Once processed, data is routed to a centralized repository for storage and analysis. This may involve a SIEM (Security Information and Event Management) offering real-time monitoring and alerting or a cloud-based data lake designed to reduce costs.
AI Response & Automation
When detection rules are triggered, or alerts are generated, SOC analysts thoroughly investigate these alerts, assess their severity levels, and implement appropriate remediation measures. Modern SOC automation solutions are evolving to adopt a more proactive approach, integrating directly with security tools rather than relying solely on SIEM alerts. This advancement allows for enhanced alert enrichment and contextual analysis, leading to more efficient remediation processes. Analysts can now differentiate real threats from false positives more quickly and conduct thorough assessments and containment strategies. For incident response, teams can now manage security incidents more effectively, conducting deeper investigations by leveraging indicators discovered during the triage phase, with legacy SOAR solutions proving particularly effective in this domain. While AI technology is poised to revolutionize SOC operations in this layer, its adoption remains complicated by SOC leaders’ ongoing concerns about SIEM-related costs and implementation challenges
Challenges With The SOC In 2025
According to survey from 300+ CISOs:
- The scale of the alert tsunami is untenable: Across the 282 organizations surveyed, teams already face ~960 security alerts every day, and enterprises with 20 k+ employees are drowning in more than 3 k alerts daily, generated by an average of 28 different tools. Analysts openly describe “way too many data sources” and a “tsunami of data” that taxes every layer of the SOC.
- Alert fatigue has become a systemic risk: On average, 40% of all alerts are simply never investigated, and 61 % of security teams admit they have ignored alerts that later proved critical, exposing customer data, taking systems offline, or driving direct business losses. Compounding the danger, the mean time to investigate sits at 70 minutes while phishing-based breaches can succeed in < 1 hour, giving adversaries a decisive head-start.
- Resource constraints are forcing teams to accept blind spots: Fifty-seven percent of organizations now suppress detection rules just to keep workloads manageable, and the first rules to be disabled are in cloud and identity, the two fastest-growing attack surfaces. Leaders acknowledge this “necessary risk,” underscoring how urgently they need automation that can restore visibility without ballooning head-count.
- Market sentiment has already shifted toward AI: “AI for Security” has vaulted into the top-three priorities for CISOs, behind only data and cloud security and 88% of organizations that do not yet run an AI-driven SOC plan to evaluate or are actively standing one up within the next 12 months. The market is moving from if to how fast AI can be integrated into everyday SOC workflows.
- AI is expected to own the majority of SOC workload inside three years: Security leaders project that AI platforms will shoulder ~60% of all SOC tasks by 2028, with 83% believing AI will handle at least half of operations. The metrics they will watch, MTTI, MTTR, and 24 ⁄ 7 coverage, align directly to AI’s strengths, signalling a near-term, ROI-driven buying cycle for purpose-built AI-SOC solutions.
AI Within The SOC
Last year, we wrote about the evolution of AI in the SOC. At the time, it was more an exploratory report into the category. In recent months, we’ve validated the problem through conversations with major security leaders globally. As SOCs grapple with escalating alert volumes, rising costs, and mounting pressure to reduce detection and response times, artificial intelligence has emerged as a necessary augmentation. It is not a replacement for human analysts, but a tool to help them manage growing operational complexity.
This second version of our report shifts the focus from exploring the need for AI in SOC operations to evaluating which AI-enabled vendor solutions are truly capable of enhancing analyst productivity, improving visibility, and reducing response times.
Recognizing ongoing skepticism around the transparency, efficacy, and adaptability of AI tools, this report segments the vendor landscape into four defined categories:
- AI-powered XDR co-pilots
- Automated Tier-1 alert analysts
- Advanced AI-driven threat hunting platforms
- Workflow-centric AI automation engineers
We apply rigorous benchmarks including measurable reductions in mean time to detect and respond (MTTD and MTTR), improvements in false-positive management, and operational cost-effectiveness to guide SOC leaders in selecting partners that align with their specific environments. This report also puts a spotlight on the use of AI agents within Security Operations Centers. SOCs typically rely on systems that take data as input and generate alerts based on predefined logic. As the volume of incoming data grows, so does the risk of alert fatigue. Agentic AI is designed to reduce that burden. By autonomously handling large volumes of alerts, it helps analysts focus on the events that truly require their judgment and expertise.
We examine how this technology is being applied across two types of environments:
- Internal SOCs operated in-house
- External SOCs managed by service providers
AI SOC Architectural Options
We reviewed many architectural breakdowns on the market, including those written by Andrew Green.
Functional Domain (What does it automate?):
Automation/Orchestration (SOAR+) & Agentic SOC
These platforms are designed to be the “central nervous system” of security operations, coordinating and automating responses across a wide range of security tools and data sources. They go beyond simple playbooks, leveraging agentic AI to intelligently sequence actions, enrich alerts, trigger containment or remediation, and handle case management, often without human intervention.Their greatest strength lies in their ability to orchestrate complex, cross-tool workflows (SIEM, EDR, cloud, ticketing, etc.) at scale, using both rules and dynamic agent logic. This results in dramatic improvements in response speed, efficiency, and consistency, particularly for large, complex environments or MSSPs.These platforms can be user-defined (built from modular blocks) or enhanced with pre-packaged agentic functions, offering the flexibility to evolve as new threats and tools emerge.
Pure-play Agentic Alert Triage Platform
While orchestration platforms focus on end-to-end workflows, this category tackles a more specific challenge: reducing the noise and burden of alert overload. These platforms rapidly triage, classify, and escalate only the most relevant threats, serving as the “first line of AI-powered defense.” They ingest high volumes of alerts from existing security systems and autonomously filter out false positives and routine events. What differentiates them is their emphasis on agentic reasoning, applying learned behaviors, contextual data, and even large language models to determine alert priority and next steps. The value is immediate: analysts are freed from “alert fatigue,” and only cases truly requiring human expertise are escalated. These are ideal for organizations looking to boost SOC productivity without a wholesale change to their architecture.
Analyst Co-Pilot/Investigation
Platforms in this domain act as “digital teammates” for human analysts, offering on-demand assistance for investigation and decision-making. Rather than automating entire workflows, these tools focus on augmenting analysts’ capabilities in real time. This can include natural language chatbots that answer questions, generate queries, summarize evidence, or suggest next steps; or more advanced reasoning engines that assemble context and walk analysts through complex incidents. What sets these apart is their role as a bridge between human expertise and machine efficiency; they’re not here to replace analysts, but to make them faster, more accurate, and less prone to error. These platforms are particularly valuable for Tier 2/3 analysts and for organizations that want to scale knowledge without losing human judgment.
Workflow/Knowledge Replication
This is the cutting edge of “institutional memory” in the SOC. Workflow/knowledge replication platforms observe, record, and learn from how the best analysts operate, then turn those behaviors into scalable, repeatable automation. Often browser-based or using workflow capture technology, these tools create digital “twins” of expert processes that can be replayed across future incidents, training new analysts and driving consistency. The unique differentiator is their ability to codify not just what to do, but how and why it’s done, preserving tacit knowledge that would otherwise be lost to turnover or scaling. They’re a powerful answer for organizations with a few superstar analysts, or for anyone seeking to operationalize best practices across distributed teams.
Implementation Model (How is it delivered?):
User-Defined/Configurable (Deterministic, agent-building, low-code)
These solutions put the power in the user’s hands: they are toolkits or platforms that let SOC teams design, customize, and continuously tune the automations, workflows, or agents that drive their security operations. Using visual interfaces, scripting, or low-code builders, users can define detection rules, orchestrate workflows, build custom agents, and adapt the platform to unique organizational requirements. This approach maximizes flexibility, adaptability, and ownership, making it ideal for organizations with mature teams or complex, evolving environments. The trade-off is that these solutions require a higher level of expertise and ongoing maintenance but the payoff is a SOC that truly fits the business’s needs.
Pre-Packaged/Black-Box (No/limited customization, R&D-driven agents)
In contrast, pre-packaged or “black-box” solutions are delivered as ready-to-run platforms with minimal end-user customization required. The underlying logic, agents, or workflows are designed and maintained by the vendor, often drawing on extensive R&D, threat intelligence, and industry best practices. This model is all about rapid time-to-value: organizations can deploy advanced AI-driven SOC capabilities quickly and easily, without the need for internal development. The trade-off is reduced flexibility; users are largely limited to the capabilities and workflows provided “out of the box.” These solutions are perfect for teams that want to modernize fast, value ease of use, or lack the bandwidth for complex customization.
Deployment Options Amongst Vendors
Beyond architecture and configuration, deployment model is another key axis of differentiation. AI SOC platforms vary in how and where they can be deployed, shaped by performance needs, regulatory requirements, and cloud readiness.The most common deployment model observed in this research was Software as a Service (SaaS), where the platform is hosted by the vendor and accessed over the internet. Some vendors also offer a “Bring Your Own Cloud” (BYOC) option, which lets clients use their own cloud infrastructure to store data and run the AI SOC platform on top of it.
Another notable deployment model, somewhat unique to AI SOC platforms, is support for air-gapped, on-premises environments. This option is particularly valuable for organizations with strict security or regulatory requirements, as it allows them to run the platform in complete isolation from external networks.
While all vendors demonstrate significant advancements in threat detection and response, the methods by which these improvements are achieved vary considerably. One of the most notable distinctions among solutions lies in their underlying architecture. Through our evaluation of multiple offerings, we identified three primary architectural approaches, each with its own advantages, limitations, and implications for how AI is deployed within the SOC.
Connected & Overlay Model On An Existing SOC (SIEM)
This model refers to AI SOC solutions that are deployed as a layer “on top of” an organization’s existing security stack. These platforms are delivered as cloud or SaaS services, and their integration into the customer environment is achieved primarily through APIs.
They do not attempt to become the central data repository or replace core SIEM/logging infrastructure. Instead, they ingest alerts and telemetry from tools like SIEM, EDR, cloud, and identity sources, then apply automated enrichment, reasoning, or response logic before handing results back to the SOC team or case management system.
Their main appeal is rapid time-to-value. Because they do not require full-scale data migration, heavy tuning, or infrastructure build-out, they can often be deployed in days or weeks. These platforms are ideal for organizations looking to enhance investigation quality, automate triage, or add a layer of AI decisioning without disrupting their existing security architecture.
The trade-off is that these solutions rely on the fidelity of alerts and data generated elsewhere, they are only as good as the signal they are fed. They also tend to have limited behavioral analytics or anomaly detection capabilities, since they rarely have access to the full raw data stream.
Examples: Prophet AI, Intezer, Dropzone AI, Radiant Security
Integrated AI SOC Platforms
These platforms take a deeper approach to integration by ingesting, storing, and analyzing security data directly. In many cases, they act as a lightweight SIEM or even a full SIEM alternative depending on the use case. Unlike overlays, they access and retain raw logs and telemetry over time, which allows for more advanced behavioral analytics and long-term anomaly detection.
The key advantage is greater visibility and analytical power. By storing data internally, these platforms can establish historical baselines, surface subtle trends, and support retrospective investigation that is not possible with overlay-only models. Many also offer cost-effective log storage and retention, which helps reduce the high costs typically associated with traditional SIEMs.
These platforms are often hybrid in nature. They can act as a log storage offload or enrichment layer for organizations with expensive or overloaded SIEMs, while also serving as standalone detection and response hubs for smaller teams.
The trade-offs include higher operational complexity, the potential for vendor lock-in since the data resides within the vendor’s environment, and additional security or compliance considerations, particularly for organizations with strict data residency or privacy requirements.
Examples: Torq, Radiant Security, Sekoia.io, D3 Security.
Human & Browser-based Workflow Emulation Platforms
This model represents the most human-centric and experiential approach. Rather than ingesting alerts through APIs and logs, these platforms capture, learn, and replicate the investigative behaviors of real analysts. They typically use browser extensions or similar technology to observe how analysts handle incidents within their native interfaces (e.g., SIEM dashboards, case management tools).
The key value lies in their ability to transform institutional knowledge and best practices into scalable, reusable automation. Over time, these platforms can “replay” these expert workflows at scale, automatically handling new incidents just as a skilled analyst would, step by step, click by click.
This approach is particularly valuable for organizations seeking to preserve and multiply the expertise of their best analysts, onboard new staff rapidly, or maintain strong consistency and quality across investigative processes.
However, there are some important caveats:
- These platforms require an upfront initial investment of time and expertise: workflows must be recorded and validated before value is realized.
- They may be slower to reach full operational impact compared to plug-and-play overlays.
- Their effectiveness depends on the presence of experienced analysts to “teach” the system
Examples: Mate, Legion Security
AI SOC vendors offer significant improvements in detection and response, but their solutions differ markedly in architectural design. Our evaluation identifies three primary architecture types, each with distinct strengths and limitations. Each architecture reflects a different approach to augmenting SOC operations. Some focus on enhancing what is already in place, while others aim to rebuild the investigative stack from the ground up. The right choice depends on an organization’s maturity, goals, and existing toolset, as well as how much control and visibility they want over their data and detection logic.
Risks and Considerations with Agentic AI in the SOC
Before diving into the methodology of this research, it is important to highlight the inherent risks of relying on Agentic AI solutions for SOC use-cases. The following considerations focus on business, operational, and compliance impacts, not technical limitations like model performance, integration effort, or feature sets.
These are the areas that leaders and decision-makers should review carefully before moving forward with agentic AI in the SOC use-cases.
- Lack of Standardized Benchmarks: There are currently no widely accepted benchmarks to evaluate agentic AI solutions in SOC environments. This makes it hard to assess performance, accuracy, or return on investment across different vendors.
- Disruption by Established Platforms: Major security vendors may integrate their own AI agents directly into existing tools, potentially reducing the need for standalone agentic AI products.
- Hype Versus Proven Impact: Agentic AI is a hot topic, but not all offerings deliver meaningful or measurable value. A cautious, evidence-based approach is needed when evaluating solutions.
- Limited Differentiation Among Vendors: Many vendors advertise the same core capabilities: triage, response, and explainability, making it difficult to distinguish real innovation from marketing noise.
- Accountability and Liability: Organizations need clear policies around responsibility and escalation when an AI system makes a wrong or harmful decision.
- Compliance with Data Regulations: Vendors must ensure that data is stored and processed in line with regional laws, including requirements around data residency and sovereignty (e.g., GDPR).
- Changing Role of the Analyst: Agentic AI shifts the SOC operating model. Analysts are moving from direct responders to overseers of automated systems, a transition that may require new skills, training, and changes to team structure.
Assessment methodology
To evaluate the capabilities of AI SOC vendors, we developed a structured assessment methodology based on real-world operational needs and expert input. Each vendor was measured across a comprehensive set of criteria that reflect the most critical functional, technical, and operational aspects of AI-driven security operations.
Our team collected data through vendor interviews, detailed questionnaires and product documentation. This information was then mapped against our evaluation matrix.
Below, we outline the key elements (not all) that make up our assessment matrix, along with explanations for why each factor is critical in evaluating the effectiveness and maturity of AI SOC platforms. These criteria reflect the capabilities we believe are most important for delivering real value in modern security operations.
1. Demonstrated efficiency improvements:
While improvements in metrics are important, the method of tracking those improvements is even more crucial. We prioritized numbers from production environments and case studies over vague, high-level statements. To demonstrate platform value, vendors should provide data points such as:
Improvement of SOC Metrics: We assessed how automation reduced the time to acknowledge and investigate alerts. Could a full investigation be completed faster with the platform than with a human analyst? By how much? Vendors were expected to support claims with real-world numbers and case studies drawn from production environments.
Alert Escalation Rate: We reviewed how many alerts were fully handled by the platform versus those that still required human escalation. This ratio served as a proxy for the level of automation maturity.
Verdict Accuracy: We evaluated how each platform measured and maintained accuracy. In particular, we looked for structured QA practices rather than simple re-testing or analyst overrides.
The rigorous quality assurance (QA) on verdict accuracy was also essential. We gave lower weight to platforms that relied on AI re-checking its own outputs or client analysts confirming/denying verdicts. Human QA teams, detection engineers, and red team exercises where the AI SOC was evaluated, were key factors for scoring well in this category.
2. Investigation speed, scalability and performance under load
This assessment category is particularly relevant for large Managed Security Service Provider (MSSP) use cases, rather than internal, well-optimized environments. We assessed on the following:
Average Investigation Time: We took into account that investigation speed can be influenced by external factors, such as the responsiveness of APIs providing contextual information. Our focus was on determining the average runtime of investigations.
Scalability: We sought to understand the approach to scalability. Was it based on a fixed value of provisioned compute resources per license, or does the architecture offer inherent scalability? We requested vendors to explain how the product performs under unusually high peak load conditions.
Performance Under Load: We inquired whether the product had undergone stress testing and if there were case studies detailing such tests or large-scale deployments.
We valued transparent explanations regarding product architecture and how scalability was achieved. We recognize that while average investigation time is a consideration, ensuring support for peak loads and the queuing of alert investigations (rather than dropping them) in resource-constrained situations, are of greater importance..
3. Context enrichment and artifact analysts
We evaluated how well the platform gathers, links and analyzes additional information (such as file behavior, user activity, threat intelligence) to provide deeper context for investigations.
Contextual data sources: We looked for whether the platform came built in with threat intelligence components such as the ability to iterate alert entities against known databases of malicious artifacts – VirtusTotal, AbuseIPDB, Recorded Future etc.
Artifact analysis: We evaluated the depth of analysis on associated artifacts. This included use of built-in sandboxing, behavioral baselining, and UEBA-style techniques.
We valued vendors who brought integrations with popular services as part of their licenses. It was important for us to understand how deep the analysis goes. We valued sandboxing, UEBA-like approach to establishing baselines of what constitutes normal behavior and analysis methods with more depth than just comparing alert entities to lists of known bad artifacts.
4. Detection of Novel Threats
We are fully aware that AI SOC platforms are not detection engines on their own. Rather, these platforms focus on analyzing already generated alerts. Nevertheless, some platforms had features that allowed them to flag novel threats.
AI evasion: We examined how the platform defended against prompt injection and other forms of AI evasion across binaries, scripts, and log repositories.
Detection of novel threats: We assessed whether the platform included any capabilities to flag anomalous behaviors or unknown patterns not already surfaced by existing tools.
We valued vendors that demonstrated thoughtful approaches to AI evasion resistance and threat novelty.
5. Depth & Breadth of Integrations
We evaluated how well each platform integrated with the broader security stack and how smoothly it fit into analyst workflows.
List of available integrations: We reviewed whether major security related solutions could be integrated out of the box. Was SOAR included as well?
Headless mode support: We assessed whether the platform could operate without requiring analysts to log into a new console,what we refer to as “headless mode.” This reflects a realistic need, as many analysts already work across multiple platforms, and adding another interface introduces friction.
Integration with communication applications: We looked for the ability to query the platform using natural language from tools like Slack or Microsoft Teams, or through built-in chatbot features.
We valued integrations into major SIEM and case management tools. Tool fatigue is a valid concern, and not having to work directly within the AI SOC platform is a welcome feature. Lack of integration with major security vendors or SOAR workflows was considered a red flag. Given how straightforward it is to connect with modern SaaS-based security solutions, we expected these integrations to be fully implemented. Integration with communication applications was also a welcome addition developed by some vendors, not essential, but certainly a strong bonus.
6. Data & Privacy
We evaluated how each platform handled data control, governance, and multi-tenant environments, all key concerns for both MSSPs and large enterprises.
Support for multi-tenant deployments: We assessed whether the platform could effectively support multiple tenants, such as MSSPs managing distinct client environments or large enterprises with segmented business units.
Control over data: We reviewed the level of control customers had over their data, including retention settings, exportability, data deletion during offboarding, and the ability to choose storage locations. We also assessed how clearly vendors articulated their data-at-rest storage practices.
Compliance with security standards. We verified alignment with regulatory and industry frameworks such as ISO 27001, SOC 2, GDPR, and NIST. Auditability and documentation were also considered.
We placed high value on flexibility and transparency in managing data. Platforms that allowed data storage in the customer’s own environment received added credit.
7. Explainability
We examined how clearly the AI system explained its actions and how effectively it incorporated analyst feedback into future decisions.
Explainability: We assessed whether every step in the AI-driven investigation was visible and auditable. This included checks for hallucination detection and clarity of decision paths.
Human feedback integration: We reviewed whether analysts could modify verdicts or influence the investigation logic, not just the final outcome. We also asked how quickly that feedback loop was implemented and how it affected future investigations.
All vendors provided some degree of step-by-step transparency. However, we prioritized platforms that enabled granular analyst input and rapid incorporation of that feedback. The more influence analysts could exert over the investigation process, the stronger the score.
Additional features
This category evaluated broader platform differentiation beyond core triage and investigation improvements. While most vendors demonstrated similar gains in Mean Time To Respond (MTTR) and investigation quality, we paid close attention to standout features and unique value propositions.
We assessed capabilities that extended beyond standard SOC workflows, such as support for compliance workflows, innovative analyst interfaces, or domain-specific use cases. These often revealed how well the platform could serve specialized teams or adapt to future needs.
We also evaluated practical operational factors like time to full deployment, vendor roadmap visibility, and customer support models. While these did not factor as heavily into the core capability scoring, they played a meaningful role in shaping the conclusions of this report.
General Market Observations
The conclusion of this assessment is not a straightforward, tiered ranking of AI SOC platforms from best to worst. Instead, we adopt a nuanced approach that integrates multiple perspectives: quantitative scoring based on our evaluation criteria, insights from vendor interviews, product demos, and broader research into the AI SOC landscape.
This multifaceted methodology is driven by two key factors:
The AI SOC market is still in its early stages: Some platforms included in our assessment have already achieved relatively wide adoption, while others remain in stealth mode. Readers should consider this variation in market maturity when interpreting our findings. Several vendors presented highly innovative concepts but lacked sufficient production deployments to validate their effectiveness with real-world case studies.
The core value proposition across AI SOC vendors is largely similar. Most platforms aim to streamline security operations by automating the investigation of alerts. In essence, they promise to reduce the volume of alerts requiring human intervention and to accelerate response times. Metrics such as Mean Time to Respond (MTTR), Mean Time to Acknowledge (MTTA), Mean Time to Detect (MTTD), or Mean Time to Investigate (MTTI) are frequently cited as key performance indicators. The ultimate goal remains the same: faster investigations and fewer alerts handled manually.
These two factors (the immaturity of the market and the shared value proposition) make direct comparative assessments challenging. However, we have developed a framework that highlights both the current capabilities of each platform and their strategic direction, based on vendor roadmaps and their broader vision for integrating AI into security operations.
In finalizing our assessment, we distilled our research into three key factors:
- Y-Axis: Overall Product Maturity
Definition: Measures the operational readiness, reliability, and trustworthiness of the platform. Evaluated based on adherence to security standards, availability and quality of support, quality assurance practices, client testimonials, case studies, and other indicators of operational robustness.
Key criteria: Security certifications, customer support quality, QA rigor, stability, real-world deployments, testimonials, case studies, length of time in market, etc.
- X-Axis: Capability Depth
Definition: Captures the range, richness, and sophistication of the platform’s features. Assessed through the breadth and sophistication of platform features, deployment flexibility, third-party integrations, roadmap commitments, handling of AI evasion techniques, ability to detect wide scope of threats and more.
Key criteria: Breadth of use cases (triage, orchestration, investigation, etc.), customization options, integrations, anomaly detection, support for novel threats, deployment options, etc.
- Quantitative Scoring:
Derived from our scoring framework across a range of criteria (some of which are detailed in this report), this factor contributes directly to the positioning of each AI SOC platform in the comparative chart below.
It is important to note that our quantitative scoring, as well as the assessments of product maturity and capability depth, do not necessarily reflect the value of each platform for your specific use case. Selecting the right AI SOC platform requires aligning your organization’s unique requirements with the broader offerings and strengths of each vendor. Below, we highlight several use cases where specific vendors demonstrated particular strengths in our assessment:
Best fit for endpoint security – Intezer
Intezer came out as one of the top AI SOC platforms in the market and an exceptionally well suited solution for environments with a strong focus on endpoint security. Originally developed as a sandboxing solution, Intezer has evolved into a comprehensive AI SOC platform, leveraging its deep roots in malware analysis and digital forensics. Its architecture, powerful built-in sandbox and a team of malware researchers, delivered exceptional results in our assessment. The platform’s endpoint-based licensing model and access to expert support further strengthen its appeal for MDR teams seeking a high-performance, analyst-friendly solution.
Most unique use-case – Legion Security, Mate Security
Legion Security and Mate Security offer one of the most distinctive use cases we have seen in this assessment. So unique, in fact, that our team had challenges comparing it to other AI SOC platforms. Legion and Mate take a fundamentally different approach. Through a browser extension, those platforms record the investigation workflows of human analysts and then automate those exact steps. This method allows organizations to effectively replicate the expertise of their best analysts at scale. What is more, those steps can then be replayed to new analysts for training purposes. According to the Legion team, future iterations may even enable importing the expertise of external analysts into the platform. Since the automation is based on real human behavior, it removes the need for human analysts to verify the accuracy of the platform’s verdicts. Although both Legion Security and Mate Security are relatively young companies, their approach addresses many of the challenges we observed in more conventional AI SOC offerings. It is an innovative solution that we’ll be watching closely as it continues to evolve.
Most unique value proposition – Radiant Security
Radiant Security stands out in the AI SOC market with one of the most compelling and value propositions we encountered. While all AI SOC platforms aim to enhance the efficiency of alert investigation at scale, Radiant goes a step further by triaging any type of alert and offering significant savings on log storage which is a major cost driver for security operations. The platform can ingest logs and compress them with approximately 70% efficiency, allowing organizations to retain logs at just 30% of their original storage footprint. These compressed logs can be stored in the customer’s own cloud environment, while Radiant provides search and investigation capabilities across that data. For some organizations, the storage cost savings alone may justify the investment in the platform. The Radiant team is clear that they are not positioning themselves as a log management solution, and their platform performed strongly across traditional AI SOC benchmarks in our assessment. However, the added benefit of reducing log storage costs sets Radiant apart in a way that is rare in today’s market.
Most unique deployment options – Crogl, Mate Security and D3 Security
The majority of AI SOC platforms follow a SaaS deployment model, requiring organizations to send sensitive data to a cloud-hosted environment. For some clients, particularly those in regulated industries or government sectors, this poses significant concerns around data sovereignty and control. Crogl, Mate Security and D3 Security directly address this challenge by offering a fully isolated, on-premises deployment option. This allows organizations to retain complete control over their data and infrastructure without sacrificing the benefits of AI-driven investigation and automation.
Great all rounders – Prophet AI, Qevlar AI, Exaforce, Dropzone AI, Command Zero
Several of the AI products we evaluated demonstrated strong, well-rounded capabilities across a range of functions. While none stood out for highly specialized or niche features, they consistently delivered robust performance in detection and response. These platforms are versatile and adaptable, making them suitable for integration into a wide variety of organizational environments with minimal friction. We recommend that readers seeking dependable, broadly effective solutions consider these offerings, especially if their priority is comprehensive coverage and reliable performance across diverse use cases.
Most feature-rich platforms – Sekoia, Torq
During our research, we observed several platforms that delivered notable improvements in detection and response. However, two platforms stood out due to their extensive feature sets, even within the AI SOC market, which already offers a wide range of robust capabilities. Sekoia.io does not yet use AI to automatically triage alerts, although this feature is on their roadmap. Nevertheless, it offers a comprehensive platform that can effectively replace a SIEM and serve as the central point for a SOC’s detection and response efforts. The platform is supported by curated threat intelligence and provides AI-powered insights that enhance the quality and speed of security investigations. Torq Hyper-SOC is not positioned as a SIEM replacement, but it provides a broad set of features including Cloud Security Posture Management, Identity and Access Management, Threat Hunting, Incident Response, Email Security, and others. Some capabilities go beyond the traditional SOC scope, such as automated onboarding and offboarding of users or triggering workflows when a cloud resource is found to be non-compliant. We feel that both of those platforms exceed expectations for features that AI SOC platforms typically bring and may be solid offerings for organizations that are looking for broad capabilities.
Top Themes Emerging from the Market
Our assessment of the AI SOC market revealed several important patterns that shaped our final conclusions. While the space is rich in innovation, many platforms converge around a similar core value proposition, with differentiation emerging only in specific design philosophies, feature sets, or architectural choices.
Converging Value Propositions Around Efficiency Throughput
Nearly all vendors in this space focus on improving the alert resolution throughput of security operations. Their core value lies in streamlining triage and investigation processes, minimizing analyst fatigue, and reducing response times. As a result, it was often challenging to distinguish vendors based solely on their core claims. Most platforms promise faster investigations, higher alert handling capacity, and reduced mean time to respond (MTTR), making marketing messages appear nearly interchangeable.
Diverging Approaches to Investigations
Where vendors began to diverge meaningfully was in how they approached the investigation process itself. We identified two distinct paradigms:
- One group of platforms constructs autonomous investigation plans, allowing AI to independently determine the investigative path and take actions based on internal logic.
- Another group uses structured guardrails, where vendors or analysts define the permissible steps and AI executes within those parameters.
This design choice reflects deeper philosophical differences regarding trust, control, and transparency. It remains to be seen which of these models will gain broader market adoption, but we view this as one of the most meaningful areas for future differentiation.
Value Beyond Investigation Throughput
Some platforms extend their value proposition beyond pure investigation throughput. For instance:
- Radiant Security introduces cost savings by offering compressed log storage that significantly reduces retention costs. Exaforce delivers similar value for log management by ingesting log data and storing it for 90 days for free.
- Legion Security and Mate emphasize trust-building by recording and replaying human-led investigations, which also supports onboarding and auditability.
- Sekoia.io and Torq position themselves as broader security automation platforms, consolidating multiple SOC functions into a single environment.
We encourage stakeholders to assess platforms across the full spectrum of capabilities and to probe for value delivered outside of traditional detection and response workflows.
Operational Guide to AI SOC Adoption
The following steps are designed to help organizations successfully integrate AI SOC products into their existing security workflows. This guide builds on our broader research to clarify which feature sets matter, identify vendors that align with specific operational needs, and provide context on where the market is heading. Some phases, particularly defining the AI strategy and selecting the right vendor, require thoughtful internal alignment and should be guided by rigorous proof-of-concept (POC) evaluations.
AI adoption is not a single decision. It is a staged process that benefits from cross-functional input, clearly defined success metrics, and a measured approach to automation. The phases below outline a practical path from early planning through full deployment, highlighting the operational checkpoints that matter most along the way.
Define the AI Security Strategy
We recommend that organizations begin by establishing a clear strategy for how AI will enhance their security operations. This should include identifying key pain points, such as alert fatigue or extended response times. AI objectives must be aligned with overarching business and security goals. Organizations should define success metrics upfront and secure stakeholder buy-in. Key questions to guide this phase include: Which SOC functions should AI improve or automate? What level of automation is appropriate? And how will success be measured?
Agree on the AI Feature Set
Next, organizations should define the capabilities required from an AI SOC solution. We recommend prioritizing core features such as threat triage and enrichment, AI-assisted investigation and response, behavioral analytics and anomaly detection, threat intelligence integration, and natural language querying. It is essential to engage stakeholders early to ensure alignment with operational needs. This is the stage where technology research and solution evaluation are most valuable.
Select an AI SOC Vendor
With requirements defined, organizations should select a vendor that aligns with their strategic goals and desired feature set. Solutions should be evaluated based on accuracy, explainability, and integration capabilities. Additional factors to consider include vendor support, regulatory compliance, and ease of deployment. Where possible, organizations should conduct pilot programs or proof-of-concept (POC) testing to validate solution effectiveness in real-world scenarios.
Deploy the AI SOC Solution
Following vendor selection, the focus should shift to integrating the AI solution into the existing environment. This includes connecting it with systems such as SIEM, EDR, SOAR, and telemetry sources. It is important to configure user roles, define operational workflows, and set up automation triggers. When available, historical data should be used to tune AI models for improved performance. Collaboration between internal teams and vendor support is critical during this phase.
Establish a Trust Period (1–2 Months)
Once deployed, we recommend an initial trust-building period focused on validating the AI system’s performance. During this time, security analysts should closely review AI-generated alerts and decisions. Feedback loops must be implemented to continuously improve accuracy. Monitoring for false positives and making configuration adjustments is essential. The goal is to build confidence in AI’s ability to support and enhance SOC operations reliably.
Transition to Full Automation
As trust in the system grows, organizations can gradually expand the level of automation. This may include automating alert triage and responding to low-risk incidents. With operational tasks increasingly handled by AI, security analysts can focus on strategic initiatives such as threat hunting, red teaming, and strengthening architecture and controls.
AI SOC can significantly enhance your organization’s ability to detect and respond to threats. However, successful adoption requires a thoughtful approach to avoid choosing the wrong solution or overestimating what AI can deliver. We recommend a careful evaluation process, particularly before enabling automated actions. Our research shows that fully autonomous operation is achievable but only after thorough due diligence and a period of trust-building between your team and the AI SOC vendor.
Conclusion
The contemporary security operations center (SOC) has reached a critical juncture, driven by increasing alert volumes, persistent staff shortages, and growing cyber threat complexity. AI-driven solutions have become essential for alleviating these operational pressures by automating alert triage, improving investigative efficiency, and enabling scalability without significant headcount increases. However, successful adoption hinges on selecting architectural models—such as Overlay, Integrated Platforms, or Human Workflow Emulation—that align closely with organizational needs, maturity, and resource availability. Additionally, the transparency and explainability of AI-driven decisions have emerged as critical factors for gaining trust and ensuring effective integration into existing workflows. While the market for AI in SOC operations is expanding rapidly, it remains relatively immature. Organizations must therefore adopt AI solutions thoughtfully, balancing anticipated benefits against potential risks including compliance, data governance, and operational disruptions.
AI technologies are increasingly vital to modern SOC operations, offering practical benefits in automation, scalability, and response speed. Despite these advantages, organizations need to carefully evaluate the suitability of different AI architectures, demand transparency from solution providers, and manage implementation risks proactively. As the AI SOC ecosystem continues to develop, informed and cautious adoption will remain essential for effectively addressing contemporary cybersecurity challenges.
