{"id":31519,"date":"2026-11-18T10:15:23","date_gmt":"2026-11-18T09:15:23","guid":{"rendered":"https:\/\/www.riskinsight-wavestone.com\/?p=31519"},"modified":"2026-11-18T10:35:09","modified_gmt":"2026-11-18T09:35:09","slug":"agentic-soc-ready-to-hand-over-the-keys-to-ai","status":"publish","type":"post","link":"https:\/\/www.riskinsight-wavestone.com\/en\/2026\/11\/agentic-soc-ready-to-hand-over-the-keys-to-ai\/","title":{"rendered":"Agentic SOC: Ready to Hand Over the Keys to AI?"},"content":{"rendered":"\n<p class=\"wp-block-paragraph\"><\/p>\n\n\n\n<p style=\"text-align: justify;\">Every day, Security Operations Centers (SOCs) face a growing volume of alerts. Most turn out to be false positives, but identifying genuine attacks means investigating each alert by correlating data across multiple tools. At scale, this workload becomes unsustainable: according to the SANS Institute (<em>2025 Detection &amp; Response Survey<\/em>), 73% of organizations rank false positives as their leading detection challenge. <strong>SOCs have no shortage of tools. What they lack is time.<\/strong><\/p>\n<p style=\"text-align: justify;\">Over the past two years, vendors have promised to solve this problem with the \u201cagentic SOC\u201d, where AI agents take on these investigations. Reality on the ground is more nuanced: in its May 2026 report, SOC-CMM, a SOC maturity framework, found that the roughly 200 SOCs surveyed still considered <strong>the value delivered by AI to be \u201climited\u201d<\/strong>.<\/p>\n<p style=\"text-align: justify;\">So, who should we believe? Based on our research, AI agents are already credible for investigations, but far less so for triaging alerts or responding on their own: their reliability in production remains to be demonstrated. For CISOs (Chief Information Security Officers), the key lies in the approach: <strong>hand over the keys one by one, use case by use case, as evidence of value and performance builds.<\/strong><\/p>\n<p>\u00a0<\/p>\n\n\n\n<h2 style=\"text-align: justify;\">From playbooks to agents: what really changes<\/h2>\n<p style=\"text-align: justify;\"><strong><em><span style=\"text-decoration: underline;\">What sets an agent apart<\/span><\/em><\/strong><\/p>\n<p style=\"text-align: justify;\">SOCs have relied on automation for years: machine learning has been used for detection for more than a decade, while SOAR (<em>Security Orchestration, Automation and Response<\/em>) <em>playbooks<\/em> automate response. <strong>A <em>playbook<\/em><\/strong> follows a predefined scenario: it enriches an alert, blocks an IP address, and so on, but <strong>stops as soon as the situation falls outside that scenario<\/strong>. More recently, <strong>AI assistants and copilots<\/strong> have helped analysts query their tools using natural language and produce reports faster, but they do not take the initiative in an investigation. Investigation, one of the most time-consuming tasks for experienced analysts, therefore remains firmly in human hands. This is where <strong>AI agents<\/strong> come in.<\/p>\n<p style=\"text-align: justify;\">Unlike a <em>playbook<\/em> or a copilot, an agent conducts the investigation itself. Typically powered by a large language model (LLM), sometimes combined with deterministic rules or machine learning, it decides, with varying degrees of freedom, what to check next. Faced with an open-ended question (\u201cIs this connection legitimate?\u201d), it breaks the problem down into individual checks, queries the SIEM (<em>Security Information and Event Management<\/em>), the EDR (<em>Endpoint Detection and Response<\/em>) or the account directory (Active Directory, for example), remembers what it has established, and changes course when a hypothesis is ruled out.<\/p>\n<p style=\"text-align: justify;\">The difference ultimately comes down to one question: who decides the next step in the investigation? In an agentic SOC, agents take on part of that decision-making. This leads us to our definition: <strong>an agentic SOC is a SOC in which AI agents handle part of the triage and investigation process, and potentially response, under the oversight of analysts.<\/strong><\/p>\n<p style=\"text-align: justify;\">The degree of freedom given to an agent can vary. In a <strong>guided approach<\/strong>, the investigation path is predefined; the LLM analyzes the results of each step, formulates hypotheses and drafts its conclusions. This sits at the boundary between traditional automation and agentic systems. In an <strong>LLM-driven approach<\/strong>, the LLM itself decides what step to take next. The former is more predictable; the latter, more flexible.<\/p>\n<p style=\"text-align: justify;\">In both cases, the level of autonomy remains a decision for each organization: just because an agent can take action does not mean it should be allowed to do so on its own. <strong>An agentic SOC is not a human-free SOC; it is a SOC in which organizations decide, task by task, which activities to delegate to agents.<\/strong><\/p>\n<p>\u00a0<\/p>\n<p style=\"text-align: justify;\"><strong><em><span style=\"text-decoration: underline;\">One alert, three ways to handle it<\/span><\/em><\/strong><\/p>\n<p style=\"text-align: justify;\">Consider a concrete example: at 3 a.m., an administrator account logs in from a country where the company has no operations, following successful multi-factor authentication (MFA).<\/p>\n<p style=\"text-align: justify;\">In a traditional SOC, the <em>playbook<\/em> enriches the case with the IP address reputation, geolocation and the account\u2019s history. Since the address is not known to be malicious, none of the <em>playbook<\/em>\u2019s conditions are met: it opens a ticket and stops there. <strong>An analyst must then conduct the investigation manually.<\/strong><\/p>\n<p style=\"text-align: justify;\">With an AI copilot, the analyst saves time: the copilot summarizes the incident, translates the analyst\u2019s questions into queries and drafts the report. But <strong>the analyst still decides which questions to ask.<\/strong><\/p>\n<p style=\"text-align: justify;\">In an agentic SOC, <strong>the agents take the lead on the investigation<\/strong>. The investigation agent considers several possible explanations (travel, VPN use or account compromise) and tests them one by one. The administrator\u2019s usual workstation is connected from France at the same time: this rules out travel. The IP address does not belong to any corporate VPN. And since MFA was successfully completed, stolen credentials alone would not explain the login: the attacker may have hijacked a session or bypassed MFA. A recently created rule forwarding emails externally from the account further strengthens the compromise hypothesis. Verdict: likely account compromise. The agents then revoke the account\u2019s active sessions, an action they are authorized to perform autonomously, and submit the deactivation of the account to the analyst for approval, as its potential impact is too significant to be decided without them.<\/p>\n<p style=\"text-align: justify;\">\u00a0<\/p>\n<figure id=\"attachment_31523\" aria-describedby=\"caption-attachment-31523\" style=\"width: 633px\" class=\"wp-caption aligncenter\"><img loading=\"lazy\" decoding=\"async\" class=\"size-full wp-image-31523\" src=\"https:\/\/www.riskinsight-wavestone.com\/wp-content\/uploads\/2026\/11\/one-alert-three-ways-to-handle-it.jpg\" alt=\"One alert three ways to handle it\" width=\"633\" height=\"356\" srcset=\"https:\/\/www.riskinsight-wavestone.com\/wp-content\/uploads\/2026\/11\/one-alert-three-ways-to-handle-it.jpg 633w, https:\/\/www.riskinsight-wavestone.com\/wp-content\/uploads\/2026\/11\/one-alert-three-ways-to-handle-it-340x191.jpg 340w, https:\/\/www.riskinsight-wavestone.com\/wp-content\/uploads\/2026\/11\/one-alert-three-ways-to-handle-it-69x39.jpg 69w\" sizes=\"auto, (max-width: 633px) 100vw, 633px\" \/><figcaption id=\"caption-attachment-31523\" class=\"wp-caption-text\">One alert three ways to handle it<\/figcaption><\/figure>\n<p style=\"text-align: justify;\">\u00a0<\/p>\n<p style=\"text-align: justify;\"><strong><em><span style=\"text-decoration: underline;\">Why now?<\/span><\/em><\/strong><\/p>\n<p style=\"text-align: justify;\">The idea of automating investigations is not new. Three developments are making it feasible today:<\/p>\n<ul style=\"text-align: justify;\">\n<li><strong>Better data quality<\/strong>: with cloud-based SIEMs and common formats such as OCSF (<em>Open Cybersecurity Schema Framework<\/em>), security logs can increasingly be queried through APIs (Application Programming Interfaces), making them directly accessible to agents.<\/li>\n<li><strong>More mature language models<\/strong>: they can use tools and perform multi-step reasoning, even though they may still produce different answers to the same question, fabricate information or be manipulated.<\/li>\n<li><strong>The rise of interoperability protocols<\/strong>: MCP (<em>Model Context Protocol<\/em>) connects agents to their tools, while A2A (<em>Agent2Agent<\/em>) aims to enable communication between agents. Many SaaS vendors have begun rolling out these new protocols.<\/li>\n<\/ul>\n<p style=\"text-align: justify;\">These protocols make it easier to connect agents without making the resulting system inherently secure: the MCP specification itself acknowledges that the protocol \u201ccannot enforce\u201d its own security principles. <strong>Agentic SOCs are therefore technically feasible, but their security and reliability still need to mature.<\/strong><\/p>\n<p>\u00a0<\/p>\n\n\n\n<h2 style=\"text-align: justify;\">What does the target agentic SOC look like?<\/h2>\n<p style=\"text-align: justify;\">The target vision presented here draws on our discussions with SOC leaders and vendors.<\/p>\n<p>\u00a0<\/p>\n<p style=\"text-align: justify;\"><strong><em><span style=\"text-decoration: underline;\">An additional layer, not an additional SOC<\/span><\/em><\/strong><\/p>\n<p style=\"text-align: justify;\">An agentic SOC does not replace the tools already in place: it adds an agentic layer between data and actions, coordinated by a <strong>supervisor agent, also known as an orchestrator<\/strong>. The analyst oversees this layer and can take control at any time.<\/p>\n<figure id=\"attachment_31525\" aria-describedby=\"caption-attachment-31525\" style=\"width: 633px\" class=\"wp-caption aligncenter\"><img loading=\"lazy\" decoding=\"async\" class=\"size-full wp-image-31525\" src=\"https:\/\/www.riskinsight-wavestone.com\/wp-content\/uploads\/2026\/11\/target-architecture-of-an-agentic-SOC.jpg\" alt=\"Target architecture of an agentic SOC\" width=\"633\" height=\"356\" srcset=\"https:\/\/www.riskinsight-wavestone.com\/wp-content\/uploads\/2026\/11\/target-architecture-of-an-agentic-SOC.jpg 633w, https:\/\/www.riskinsight-wavestone.com\/wp-content\/uploads\/2026\/11\/target-architecture-of-an-agentic-SOC-340x191.jpg 340w, https:\/\/www.riskinsight-wavestone.com\/wp-content\/uploads\/2026\/11\/target-architecture-of-an-agentic-SOC-69x39.jpg 69w\" sizes=\"auto, (max-width: 633px) 100vw, 633px\" \/><figcaption id=\"caption-attachment-31525\" class=\"wp-caption-text\">Target architecture of an agentic SOC<\/figcaption><\/figure>\n<p style=\"text-align: justify;\">\u00a0<\/p>\n<p style=\"text-align: justify;\">Existing tools (SIEM, EDR, SOAR, etc.) form the foundation: they provide data that has already been normalized, correlated and enriched. Agents connect to them through APIs or MCP servers, first to retrieve this data and its context, then to take action, either directly or through SOAR <em>playbooks<\/em>.<\/p>\n<p style=\"text-align: justify;\">The agentic layer brings together <strong>four roles<\/strong>, each performed by a specialized agent: the triage agent qualifies the alert; the investigation agent investigates it through to a verdict; the enrichment agent provides threat intelligence; and the response agent prepares the actions to be taken on the IT environment. These agents may be separate or combined into a single agent; we always refer to them by their role.<\/p>\n<p style=\"text-align: justify;\">The <strong>supervisor agent<\/strong> coordinates the overall process: typically powered by an LLM, it breaks the alert down into tasks, assigns them to specialized agents, monitors their execution and triggers the actions prepared by the response agent, within its authorized scope. It coordinates the agents, but does not have the final say: the analyst remains in control. In the remainder of this article, \u201cthe agents\u201d refers to the entire agentic layer; where a point relates to a specific agent, we name it explicitly.<\/p>\n<p style=\"text-align: justify;\">An <strong>investigation memory<\/strong>, shared by all agents, stores hypotheses, evidence and verdicts, confirmed or corrected by analysts, and provides context for future investigations.<\/p>\n<p style=\"text-align: justify;\">Finally, <strong>cross-cutting controls<\/strong> govern the entire system: least-privilege permissions separating read access from action permissions, input filtering, traceability of every query and decision, and human approval for sensitive actions.<\/p>\n<p style=\"text-align: justify;\">Input filtering addresses a risk specific to language models: prompt injection. An attacker can embed instructions in a username, URL or log field to manipulate an agent\u2019s reasoning. According to a May 2026 preprint that has not been peer reviewed, such instructions can conceal 68% of malicious logs from an unprotected analysis agent. <strong>Any information must therefore be treated as potentially malicious.<\/strong><\/p>\n<p>\u00a0<\/p>\n<p style=\"text-align: justify;\"><strong><em><span style=\"text-decoration: underline;\">The alert lifecycle: what agents do and what remains with the analyst<\/span><\/em><\/strong><\/p>\n<p style=\"text-align: justify;\">The stages involved in handling an alert do not change: detect, triage, investigate, decide, respond and learn. What does change is <strong>how the work is divided between agents and analysts<\/strong>.<\/p>\n<p style=\"text-align: justify;\">At the end of an investigation, the investigation agent produces a reasoned verdict accompanied by a confidence score. There are three possible outcomes: a false positive is closed and documented; an inconclusive case is escalated to the analyst along with the avenues already explored; a confirmed or probable threat is passed to the response agent, which prepares the appropriate actions. The supervisor agent automatically triggers actions that have been pre-authorized and submits all others to the analyst.<\/p>\n<p style=\"text-align: justify;\">The confidence score can be used to adjust this autonomy, <strong>provided that it has been calibrated against real-world cases<\/strong>: a score of 90% is meaningful only if 90% of verdicts assigned that score are actually correct. In all cases, <strong>analysts retain control over decisions that materially affect the organization<\/strong>: ambiguous cases, sensitive actions and detection tuning.<\/p>\n<p style=\"text-align: justify;\">Most of the time saved often comes from closing false positives: in a randomized trial conducted by Microsoft on the triage of phishing reports, around 83% of the time savings came from reports classified as benign. This is also where mistakes are most costly: <strong>an alert incorrectly classified as benign is an attack that goes unnoticed.<\/strong> Automated closures should therefore be audited through sampling.<\/p>\n<p style=\"text-align: justify;\">Finally, learning remains deliberately limited: agents do not retrain their underlying models; instead, they enrich the investigation memory using analysts\u2019 corrections. <strong>Every learned element thus remains visible and reversible.<\/strong><\/p>\n<p style=\"text-align: justify;\"><strong>What happens to Tier 1 analysts?<\/strong> Their role does not disappear, but it changes: less repetitive alert qualification, more auditing of automated closures, reviewing investigation cases and correcting agent verdicts. Training analysts directly for Tier 2 roles may be tempting, but it is risky: handling alerts is how analysts learn to recognize an attack. <strong>Tier 1 becomes a supervised apprenticeship role rather than a production role.<\/strong><\/p>\n<p>\u00a0<\/p>\n\n\n\n<h2 style=\"text-align: justify;\">The agentic SOC market: where do things stand?<\/h2>\n<p style=\"text-align: justify;\"><strong><em><span style=\"text-decoration: underline;\">A young, fast-moving market<\/span><\/em><\/strong><\/p>\n<p style=\"text-align: justify;\">The agentic SOC is no longer just a trade-show concept: we identified sixteen vendors, divided into two broad categories. On one side, <strong>security platforms<\/strong> (SIEM, EDR or XDR, for <em>Extended Detection and Response<\/em>) are adding agents to products that already have access to security data and enforcement mechanisms. On the other, <strong>specialized vendors<\/strong>, often younger companies, offer \u201cAI SOC analysts\u201d that connect to customers\u2019 existing tools.<\/p>\n<p>\u00a0<\/p>\n<p style=\"text-align: justify;\"><strong><em><span style=\"text-decoration: underline;\">Our methodology: five solutions under the microscope<\/span><\/em><\/strong><\/p>\n<p style=\"text-align: justify;\">We conducted an in-depth assessment of <strong>five solutions representing both categories<\/strong>, using 26 criteria derived from RFPs, discussions with SOC leaders and workshops with vendors. The criteria are grouped into seven areas: detection, triage, investigation, response and human validation, reliability of conclusions, agent coordination, and governance. Each criterion is assessed against a predefined scale, with the highest levels requiring measured results. This approach has its limitations: it is no substitute for testing solutions against a dataset of alerts, does not cover the entire market and reflects the state of the solutions at the time of our study. We therefore present trends rather than a ranking.<\/p>\n<p>\u00a0<\/p>\n<p style=\"text-align: justify;\"><strong><em><span style=\"text-decoration: underline;\">Platform or standalone solution: the first decision to make<\/span><\/em><\/strong><\/p>\n<p style=\"text-align: justify;\"><strong>Where are the agents integrated?<\/strong> The integration point determines which data the agents can see and how their actions are executed. Two models coexist, broadly corresponding to the two categories of vendors.<\/p>\n<figure id=\"attachment_31527\" aria-describedby=\"caption-attachment-31527\" style=\"width: 633px\" class=\"wp-caption aligncenter\"><img loading=\"lazy\" decoding=\"async\" class=\"size-full wp-image-31527\" src=\"https:\/\/www.riskinsight-wavestone.com\/wp-content\/uploads\/2026\/11\/two-models-for-integrating-agents.png\" alt=\"Two models for integrating agents\" width=\"633\" height=\"356\" srcset=\"https:\/\/www.riskinsight-wavestone.com\/wp-content\/uploads\/2026\/11\/two-models-for-integrating-agents.png 633w, https:\/\/www.riskinsight-wavestone.com\/wp-content\/uploads\/2026\/11\/two-models-for-integrating-agents-340x191.png 340w, https:\/\/www.riskinsight-wavestone.com\/wp-content\/uploads\/2026\/11\/two-models-for-integrating-agents-69x39.png 69w\" sizes=\"auto, (max-width: 633px) 100vw, 633px\" \/><figcaption id=\"caption-attachment-31527\" class=\"wp-caption-text\">Two models for integrating agents<\/figcaption><\/figure>\n<p style=\"text-align: justify;\">\u00a0<\/p>\n<p style=\"text-align: justify;\">In the first model, <strong>agents are integrated into a security platform<\/strong>, which also provides the supervisor agent. They leverage data from the platform\u2019s sensors, already normalized and correlated, and act through its native capabilities. Its strength is integration: a guarantee of consistency and speed for a SOC built around the platform, which also benefits from the vendor\u2019s new capabilities as they are released. The trade-off is dependency: agents primarily see what the platform sees, and their coverage of third-party tools must be verified.<\/p>\n<p style=\"text-align: justify;\">In the second model, <strong>a standalone solution, with its own supervisor agent and specialized agents, connects to the existing tools<\/strong>. Without its own sensors or native response capabilities, and usually without its own database, it queries and operates the customer\u2019s tools through APIs, or forwards its recommendations to the SOAR or the analyst. Its strength is adaptability: it is well suited to heterogeneous environments and multi-client service providers. But <strong>its effectiveness depends entirely on the quality of data and connectors<\/strong>. And as the market consolidates, with large platforms acquiring specialist vendors, a vendor\u2019s long-term viability matters just as much as its technology.<\/p>\n<p style=\"text-align: justify;\">The choice therefore comes down to a trade-off: <strong>on one side, the native integration and ease of evolution offered by a platform\u2019s ecosystem; on the other, the ability to adapt to heterogeneous environments.<\/strong> These acquisitions are, however, beginning to blur the boundary between the two models.<\/p>\n<p>\u00a0<\/p>\n<p style=\"text-align: justify;\"><strong><em><span style=\"text-decoration: underline;\">What is mature, and what remains to be proven<\/span><\/em><\/strong><\/p>\n<p style=\"text-align: justify;\">Not every stage of the alert lifecycle has reached the same point. Our analysis identifies three levels of maturity:<\/p>\n<ul style=\"text-align: justify;\">\n<li><strong>Most advanced: investigation and enrichment.<\/strong> All solutions in our sample claim the same capabilities: building an investigation plan, including for alerts no <em>playbook<\/em> anticipated; chaining queries across tools; reconstructing an attack timeline; providing threat intelligence; and producing a reasoned verdict that the analyst can correct. Because these steps do not modify the IT environment, mistakes remain recoverable.<\/li>\n<li><strong>Promising, but requiring production validation: triage and response.<\/strong> The capabilities exist, but vary significantly across solutions and involve decisions that commit the organization: closing an alert, acting on the IT environment. Errors are more costly: an alert closed incorrectly lets an attack through, and a poorly targeted action disrupts business operations.<\/li>\n<li><strong>Still emerging: agentic detection.<\/strong> Machine-learning-based detection has been mature for years; what remains emerging is an agent capable of creating or adjusting detection rules on its own. Few solutions in our sample offer this, and the rules generated are generally still validated by an analyst before going into production. ANSSI likewise classifies automated rule generation among the most difficult use cases to implement. The SIEM is therefore not dead. Especially since no LLM could ingest everything it sees at a reasonable cost: agents rely on it to collect, correlate and filter data, then investigate what it surfaces.<\/li>\n<\/ul>\n<p>\u00a0<\/p>\n<p style=\"text-align: justify;\"><strong><em><span style=\"text-decoration: underline;\">Autonomy: the real differentiator<\/span><\/em><\/strong><\/p>\n<p style=\"text-align: justify;\">When it comes to closing alerts or acting on the IT environment, solutions diverge: some only make recommendations, while others act directly on endpoints, identities, networks or cloud environments. The same \u201cAI SOC\u201d label thus covers simple assistants, tools limited to triage and solutions covering the entire lifecycle. <strong>Two solutions marketed as \u201cautonomous\u201d may therefore offer very different capabilities.<\/strong><\/p>\n<p style=\"text-align: justify;\">The safeguards governing this autonomy vary just as widely. Depending on the solution, everything goes through the analyst, agents act on their own above a given confidence threshold, or a matrix defines precisely what each agent is allowed to do, and on which resources. Yet an AI cannot assess its own reliability with certainty: its confidence score is only meaningful if it has been calibrated against verdicts established by analysts. <strong>None of the solutions in our sample publishes such a calibration<\/strong>, even though this autonomy often depends on that very score.<\/p>\n<p>\u00a0<\/p>\n<p style=\"text-align: justify;\"><strong><em><span style=\"text-decoration: underline;\">The missing piece: real-world evidence<\/span><\/em><\/strong><\/p>\n<p style=\"text-align: justify;\"><strong>On paper, the capabilities are there; they now need to be validated in real-world conditions, particularly for triage and response.<\/strong> No reliability measurements have been published on a real volume of alerts. Three questions remain open:<\/p>\n<ul style=\"text-align: justify;\">\n<li><strong>Reproducibility<\/strong>: when presented with the same alert, does the agent consistently reach the same conclusion? ANSSI describes this as a \u201cmajor challenge\u201d, yet no metric is available to assess it.<\/li>\n<li><strong>Comparability<\/strong>: there is no reference dataset of alerts against which solutions can be compared, and \u201cautonomous closure\u201d does not mean the same thing from one product to another.<\/li>\n<li><strong>Benefits<\/strong>: they are mostly vendor claims, often expressed as \u201cup to X%\u201d, and rarely measured in production.<\/li>\n<\/ul>\n<p style=\"text-align: justify;\">Evidence therefore needs to come from both sides: vendors need to publish comparable results, and customers need to test solutions against their own alerts. This is precisely the purpose of our recommendations.<\/p>\n<p>\u00a0<\/p>\n\n\n\n<h2 style=\"text-align: justify;\">Our recommendations for CISOs<\/h2>\n<p style=\"text-align: justify;\"><strong>The market is mature enough to get started, but not yet mature enough for uncontrolled delegation.<\/strong> Waiting for it to stabilize would mean missing out on the experience organizations will need to entrust more to agents tomorrow.<\/p>\n<p>\u00a0<\/p>\n<p style=\"text-align: justify;\"><strong><em><span style=\"text-decoration: underline;\">Set the rules first<\/span><\/em><\/strong><\/p>\n<p style=\"text-align: justify;\"><strong>Establish the governance rules before granting any autonomy<\/strong>: give each agent its own identity and least-privilege permissions, strictly separate read and action privileges, log every query and decision, and treat any information provided to agents as potentially malicious. The OWASP Top 10 for Agentic Applications, published in December 2025, provides a good starting framework. Then define, action by action, what agents may trigger on their own, based on the impact of the action and how easily it can be reversed, and submit everything else for human approval. This delegation should be expanded step by step, as evidence builds. These rules also support compliance with the European NIS2 Directive and, for the financial sector, the DORA Regulation.<\/p>\n<p>\u00a0<\/p>\n<p style=\"text-align: justify;\"><strong><em><span style=\"text-decoration: underline;\">Start small, measure everything<\/span><\/em><\/strong><\/p>\n<p style=\"text-align: justify;\">Next, <strong>select use cases where agents deliver value without disproportionate risk<\/strong>, and open them up in the following order:<\/p>\n<ul style=\"text-align: justify;\">\n<li>investigation and enrichment, as good starting points;<\/li>\n<li>triage next, with close monitoring of false negatives, meaning attacks incorrectly classified as benign;<\/li>\n<li>response last, beginning with low-impact actions that are easy to reverse.<\/li>\n<\/ul>\n<p style=\"text-align: justify;\">For an initial pilot, investigating phishing reports, identity-related alerts and EDR alerts provides a good testing ground.<\/p>\n<p style=\"text-align: justify;\">The simplest approach is to start with the agentic capabilities already built into the SOC\u2019s tools. To go further, choose the integration model before the product; building your own agents only makes sense for highly specific data sources, such as business applications.<\/p>\n<p style=\"text-align: justify;\">When dealing with vendors, demand evidence: is the feature generally available or still in preview? Has it been measured in production? What exactly does the advertised \u201cautonomous closure\u201d cover?<\/p>\n<p style=\"text-align: justify;\">Start the pilot in observation mode: agents propose, analysts decide, and agent verdicts are compared with those of analysts. Above all, <strong>make sure it leads to a business case<\/strong>: for which use cases, and from what volume, do agents actually create value? In practical terms:<\/p>\n<ul style=\"text-align: justify;\">\n<li>build a dataset of previously handled alerts with known verdicts, to compare solutions on your own data;<\/li>\n<li>define metrics in advance (time to triage, escalation rate to analysts, automated closure rate, false negatives measured through sampling, rejection rate of agent recommendations), together with the thresholds that must be met before autonomy can be expanded;<\/li>\n<li>compare the cost of handling an alert with agents, based on the vendor\u2019s pricing model (per alert, by volume or per license), against the current cost;<\/li>\n<li>test what vendors have yet to demonstrate: reproducibility of conclusions on replayed alerts, reliability of the confidence score, behavior when data is missing, the ability to return an \u201cinconclusive\u201d verdict, and resilience against prompt injection in the data provided to agents.<\/li>\n<\/ul>\n<p>\u00a0<\/p>\n<p style=\"text-align: justify;\"><strong><em><span style=\"text-decoration: underline;\">Prepare your teams<\/span><\/em><\/strong><\/p>\n<p style=\"text-align: justify;\">Supervising agents is a skill in its own right: reviewing an investigation case produced by an agent, identifying its mistakes and resisting automation bias (the tendency to follow a machine\u2019s recommendation without verifying it). For new analysts, this means replaying previous investigations, reviewing those conducted by agents and practicing on simulated cases. <strong>Training pathways need to be redesigned accordingly.<\/strong><\/p>\n<p>\u00a0<\/p>\n\n\n\n<p style=\"text-align: justify;\">The agentic SOC offering is expanding rapidly, and agent-led investigation already delivers on its promises on paper. <strong>Its reliability in production, however, remains to be demonstrated<\/strong>, and the non-deterministic nature of agent reasoning calls for safeguards that traditional automation did not require.<\/p>\n<p style=\"text-align: justify;\">Today, <strong>investigation is the right place to start<\/strong>: this is where agents are most mature and where mistakes remain recoverable. From there, <strong>autonomy must be earned, not granted by default<\/strong>: it should gradually extend to triage and then response as evidence builds on the organization\u2019s own alerts. Finally, <strong>analysts are not disappearing; their role is changing<\/strong>: less gathering of information, more oversight of decisions.<\/p>\n<p style=\"text-align: justify;\">To the question \u201cWho decides the next step in the investigation?\u201d, the answer will therefore be a graduated one: agents where their reliability has been proven and the risk is under control; analysts everywhere else. So, ready to hand over the keys to AI? <strong>Not all of them yet, and not all at once: handing them over one by one, as evidence of value and performance builds, will be as much the responsibility of CISOs as of vendors.<\/strong><\/p>\n","protected":false},"excerpt":{"rendered":"<p>Every day, Security Operations Centers (SOCs) face a growing volume of alerts. Most turn out to be false positives, but identifying genuine attacks means investigating each alert by correlating data across multiple tools. At scale, this workload becomes unsustainable: according&#8230;<\/p>\n","protected":false},"author":92,"featured_media":31537,"comment_status":"open","ping_status":"closed","sticky":false,"template":"page-templates\/tmpl-one.php","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[3270,2777,3977],"tags":[5230,5004,3519,3422,4297,5231,5035,3126],"coauthors":[931,5227,4894],"class_list":["post-31519","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-cyberrisk-management-strategy-en","category-cybersecurity-digital-trust","category-focus","tag-agentic-soc","tag-ai-agents","tag-ciso","tag-detection-en-2","tag-llm-2","tag-mcp","tag-siem","tag-soc-en"],"acf":[],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v28.1 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>Agentic SOC: Ready to Hand Over the Keys to AI? - RiskInsight<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.riskinsight-wavestone.com\/en\/2026\/11\/agentic-soc-ready-to-hand-over-the-keys-to-ai\/\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Agentic SOC: Ready to Hand Over the Keys to AI? - RiskInsight\" \/>\n<meta property=\"og:description\" content=\"Every day, Security Operations Centers (SOCs) face a growing volume of alerts. Most turn out to be false positives, but identifying genuine attacks means investigating each alert by correlating data across multiple tools. At scale, this workload becomes unsustainable: according...\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.riskinsight-wavestone.com\/en\/2026\/11\/agentic-soc-ready-to-hand-over-the-keys-to-ai\/\" \/>\n<meta property=\"og:site_name\" content=\"RiskInsight\" \/>\n<meta property=\"article:published_time\" content=\"2026-11-18T09:15:23+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2026-11-18T09:35:09+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.riskinsight-wavestone.com\/wp-content\/uploads\/2026\/11\/cover2.png\" \/>\n\t<meta property=\"og:image:width\" content=\"1341\" \/>\n\t<meta property=\"og:image:height\" content=\"894\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/png\" \/>\n<meta name=\"author\" content=\"Beno\u00eet Marion, Jason Lau, Gael Audinet\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"Beno\u00eet Marion, Jason Lau, Gael Audinet\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"19 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/en\\\/2026\\\/11\\\/agentic-soc-ready-to-hand-over-the-keys-to-ai\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/en\\\/2026\\\/11\\\/agentic-soc-ready-to-hand-over-the-keys-to-ai\\\/\"},\"author\":{\"name\":\"Beno\u00eet Marion\",\"@id\":\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/en\\\/#\\\/schema\\\/person\\\/8269382799f649bb774493a375d57d31\"},\"headline\":\"Agentic SOC: Ready to Hand Over the Keys to AI?\",\"datePublished\":\"2026-11-18T09:15:23+00:00\",\"dateModified\":\"2026-11-18T09:35:09+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/en\\\/2026\\\/11\\\/agentic-soc-ready-to-hand-over-the-keys-to-ai\\\/\"},\"wordCount\":3481,\"commentCount\":0,\"publisher\":{\"@id\":\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/en\\\/#organization\"},\"image\":{\"@id\":\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/en\\\/2026\\\/11\\\/agentic-soc-ready-to-hand-over-the-keys-to-ai\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/wp-content\\\/uploads\\\/2026\\\/11\\\/cover2.png\",\"keywords\":[\"Agentic SOC\",\"AI agents\",\"CISO\",\"detection\",\"LLM\",\"MCP\",\"SIEM\",\"SOC\"],\"articleSection\":[\"Cyberrisk Management &amp; Strategy\",\"Cybersecurity &amp; Digital Trust\",\"Focus\"],\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/en\\\/2026\\\/11\\\/agentic-soc-ready-to-hand-over-the-keys-to-ai\\\/#respond\"]}]},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/en\\\/2026\\\/11\\\/agentic-soc-ready-to-hand-over-the-keys-to-ai\\\/\",\"url\":\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/en\\\/2026\\\/11\\\/agentic-soc-ready-to-hand-over-the-keys-to-ai\\\/\",\"name\":\"Agentic SOC: Ready to Hand Over the Keys to AI? - RiskInsight\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/en\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/en\\\/2026\\\/11\\\/agentic-soc-ready-to-hand-over-the-keys-to-ai\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/en\\\/2026\\\/11\\\/agentic-soc-ready-to-hand-over-the-keys-to-ai\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/wp-content\\\/uploads\\\/2026\\\/11\\\/cover2.png\",\"datePublished\":\"2026-11-18T09:15:23+00:00\",\"dateModified\":\"2026-11-18T09:35:09+00:00\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/en\\\/2026\\\/11\\\/agentic-soc-ready-to-hand-over-the-keys-to-ai\\\/#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/en\\\/2026\\\/11\\\/agentic-soc-ready-to-hand-over-the-keys-to-ai\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/en\\\/2026\\\/11\\\/agentic-soc-ready-to-hand-over-the-keys-to-ai\\\/#primaryimage\",\"url\":\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/wp-content\\\/uploads\\\/2026\\\/11\\\/cover2.png\",\"contentUrl\":\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/wp-content\\\/uploads\\\/2026\\\/11\\\/cover2.png\",\"width\":1341,\"height\":894},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/en\\\/2026\\\/11\\\/agentic-soc-ready-to-hand-over-the-keys-to-ai\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Accueil\",\"item\":\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/en\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Agentic SOC: Ready to Hand Over the Keys to AI?\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/en\\\/#website\",\"url\":\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/en\\\/\",\"name\":\"RiskInsight\",\"description\":\"The cybersecurity &amp; digital trust blog by Wavestone&#039;s consultants\",\"publisher\":{\"@id\":\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/en\\\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/en\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-US\"},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/en\\\/#organization\",\"name\":\"Wavestone\",\"url\":\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/en\\\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/en\\\/#\\\/schema\\\/logo\\\/image\\\/\",\"url\":\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/wp-content\\\/uploads\\\/2021\\\/08\\\/Monogramme\u2013W\u2013NEGA-RGB-50x50-1.png\",\"contentUrl\":\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/wp-content\\\/uploads\\\/2021\\\/08\\\/Monogramme\u2013W\u2013NEGA-RGB-50x50-1.png\",\"width\":50,\"height\":50,\"caption\":\"Wavestone\"},\"image\":{\"@id\":\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/en\\\/#\\\/schema\\\/logo\\\/image\\\/\"}},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/en\\\/#\\\/schema\\\/person\\\/8269382799f649bb774493a375d57d31\",\"name\":\"Beno\u00eet Marion\",\"url\":\"https:\\\/\\\/www.riskinsight-wavestone.com\\\/en\\\/author\\\/benoit-marion\\\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Agentic SOC: Ready to Hand Over the Keys to AI? - RiskInsight","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.riskinsight-wavestone.com\/en\/2026\/11\/agentic-soc-ready-to-hand-over-the-keys-to-ai\/","og_locale":"en_US","og_type":"article","og_title":"Agentic SOC: Ready to Hand Over the Keys to AI? - RiskInsight","og_description":"Every day, Security Operations Centers (SOCs) face a growing volume of alerts. Most turn out to be false positives, but identifying genuine attacks means investigating each alert by correlating data across multiple tools. At scale, this workload becomes unsustainable: according...","og_url":"https:\/\/www.riskinsight-wavestone.com\/en\/2026\/11\/agentic-soc-ready-to-hand-over-the-keys-to-ai\/","og_site_name":"RiskInsight","article_published_time":"2026-11-18T09:15:23+00:00","article_modified_time":"2026-11-18T09:35:09+00:00","og_image":[{"width":1341,"height":894,"url":"https:\/\/www.riskinsight-wavestone.com\/wp-content\/uploads\/2026\/11\/cover2.png","type":"image\/png"}],"author":"Beno\u00eet Marion, Jason Lau, Gael Audinet","twitter_misc":{"Written by":"Beno\u00eet Marion, Jason Lau, Gael Audinet","Est. reading time":"19 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/www.riskinsight-wavestone.com\/en\/2026\/11\/agentic-soc-ready-to-hand-over-the-keys-to-ai\/#article","isPartOf":{"@id":"https:\/\/www.riskinsight-wavestone.com\/en\/2026\/11\/agentic-soc-ready-to-hand-over-the-keys-to-ai\/"},"author":{"name":"Beno\u00eet Marion","@id":"https:\/\/www.riskinsight-wavestone.com\/en\/#\/schema\/person\/8269382799f649bb774493a375d57d31"},"headline":"Agentic SOC: Ready to Hand Over the Keys to AI?","datePublished":"2026-11-18T09:15:23+00:00","dateModified":"2026-11-18T09:35:09+00:00","mainEntityOfPage":{"@id":"https:\/\/www.riskinsight-wavestone.com\/en\/2026\/11\/agentic-soc-ready-to-hand-over-the-keys-to-ai\/"},"wordCount":3481,"commentCount":0,"publisher":{"@id":"https:\/\/www.riskinsight-wavestone.com\/en\/#organization"},"image":{"@id":"https:\/\/www.riskinsight-wavestone.com\/en\/2026\/11\/agentic-soc-ready-to-hand-over-the-keys-to-ai\/#primaryimage"},"thumbnailUrl":"https:\/\/www.riskinsight-wavestone.com\/wp-content\/uploads\/2026\/11\/cover2.png","keywords":["Agentic SOC","AI agents","CISO","detection","LLM","MCP","SIEM","SOC"],"articleSection":["Cyberrisk Management &amp; Strategy","Cybersecurity &amp; Digital Trust","Focus"],"inLanguage":"en-US","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/www.riskinsight-wavestone.com\/en\/2026\/11\/agentic-soc-ready-to-hand-over-the-keys-to-ai\/#respond"]}]},{"@type":"WebPage","@id":"https:\/\/www.riskinsight-wavestone.com\/en\/2026\/11\/agentic-soc-ready-to-hand-over-the-keys-to-ai\/","url":"https:\/\/www.riskinsight-wavestone.com\/en\/2026\/11\/agentic-soc-ready-to-hand-over-the-keys-to-ai\/","name":"Agentic SOC: Ready to Hand Over the Keys to AI? - RiskInsight","isPartOf":{"@id":"https:\/\/www.riskinsight-wavestone.com\/en\/#website"},"primaryImageOfPage":{"@id":"https:\/\/www.riskinsight-wavestone.com\/en\/2026\/11\/agentic-soc-ready-to-hand-over-the-keys-to-ai\/#primaryimage"},"image":{"@id":"https:\/\/www.riskinsight-wavestone.com\/en\/2026\/11\/agentic-soc-ready-to-hand-over-the-keys-to-ai\/#primaryimage"},"thumbnailUrl":"https:\/\/www.riskinsight-wavestone.com\/wp-content\/uploads\/2026\/11\/cover2.png","datePublished":"2026-11-18T09:15:23+00:00","dateModified":"2026-11-18T09:35:09+00:00","breadcrumb":{"@id":"https:\/\/www.riskinsight-wavestone.com\/en\/2026\/11\/agentic-soc-ready-to-hand-over-the-keys-to-ai\/#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.riskinsight-wavestone.com\/en\/2026\/11\/agentic-soc-ready-to-hand-over-the-keys-to-ai\/"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/www.riskinsight-wavestone.com\/en\/2026\/11\/agentic-soc-ready-to-hand-over-the-keys-to-ai\/#primaryimage","url":"https:\/\/www.riskinsight-wavestone.com\/wp-content\/uploads\/2026\/11\/cover2.png","contentUrl":"https:\/\/www.riskinsight-wavestone.com\/wp-content\/uploads\/2026\/11\/cover2.png","width":1341,"height":894},{"@type":"BreadcrumbList","@id":"https:\/\/www.riskinsight-wavestone.com\/en\/2026\/11\/agentic-soc-ready-to-hand-over-the-keys-to-ai\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Accueil","item":"https:\/\/www.riskinsight-wavestone.com\/en\/"},{"@type":"ListItem","position":2,"name":"Agentic SOC: Ready to Hand Over the Keys to AI?"}]},{"@type":"WebSite","@id":"https:\/\/www.riskinsight-wavestone.com\/en\/#website","url":"https:\/\/www.riskinsight-wavestone.com\/en\/","name":"RiskInsight","description":"The cybersecurity &amp; digital trust blog by Wavestone&#039;s consultants","publisher":{"@id":"https:\/\/www.riskinsight-wavestone.com\/en\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.riskinsight-wavestone.com\/en\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"},{"@type":"Organization","@id":"https:\/\/www.riskinsight-wavestone.com\/en\/#organization","name":"Wavestone","url":"https:\/\/www.riskinsight-wavestone.com\/en\/","logo":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/www.riskinsight-wavestone.com\/en\/#\/schema\/logo\/image\/","url":"https:\/\/www.riskinsight-wavestone.com\/wp-content\/uploads\/2021\/08\/Monogramme\u2013W\u2013NEGA-RGB-50x50-1.png","contentUrl":"https:\/\/www.riskinsight-wavestone.com\/wp-content\/uploads\/2021\/08\/Monogramme\u2013W\u2013NEGA-RGB-50x50-1.png","width":50,"height":50,"caption":"Wavestone"},"image":{"@id":"https:\/\/www.riskinsight-wavestone.com\/en\/#\/schema\/logo\/image\/"}},{"@type":"Person","@id":"https:\/\/www.riskinsight-wavestone.com\/en\/#\/schema\/person\/8269382799f649bb774493a375d57d31","name":"Beno\u00eet Marion","url":"https:\/\/www.riskinsight-wavestone.com\/en\/author\/benoit-marion\/"}]}},"_links":{"self":[{"href":"https:\/\/www.riskinsight-wavestone.com\/en\/wp-json\/wp\/v2\/posts\/31519","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.riskinsight-wavestone.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.riskinsight-wavestone.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.riskinsight-wavestone.com\/en\/wp-json\/wp\/v2\/users\/92"}],"replies":[{"embeddable":true,"href":"https:\/\/www.riskinsight-wavestone.com\/en\/wp-json\/wp\/v2\/comments?post=31519"}],"version-history":[{"count":4,"href":"https:\/\/www.riskinsight-wavestone.com\/en\/wp-json\/wp\/v2\/posts\/31519\/revisions"}],"predecessor-version":[{"id":31530,"href":"https:\/\/www.riskinsight-wavestone.com\/en\/wp-json\/wp\/v2\/posts\/31519\/revisions\/31530"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.riskinsight-wavestone.com\/en\/wp-json\/wp\/v2\/media\/31537"}],"wp:attachment":[{"href":"https:\/\/www.riskinsight-wavestone.com\/en\/wp-json\/wp\/v2\/media?parent=31519"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.riskinsight-wavestone.com\/en\/wp-json\/wp\/v2\/categories?post=31519"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.riskinsight-wavestone.com\/en\/wp-json\/wp\/v2\/tags?post=31519"},{"taxonomy":"author","embeddable":true,"href":"https:\/\/www.riskinsight-wavestone.com\/en\/wp-json\/wp\/v2\/coauthors?post=31519"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}