agentic security
1 TopicWhat's new with Microsoft Perception at Ignite 2026
A year ago, a campaign powered by EvilTokens, a phishing-as-a-service kit, took roughly fourteen days to run its course. More recently, comparable activity observed by Microsoft Threat Intelligence unfolded in just 85 minutes and expanded from 10 targeted users to 3,000. Attackers are using AI to move faster, scale operations, and execute attacks in parallel. But defense still too often happens one alert, one investigation, and one response at a time. What if defense could work differently? What if AI agents could anticipate threats, investigate attacks, and take coordinated action across your environment, instead of waiting for the next alert? Microsoft Perception is built for this new reality. At Microsoft Ignite, Project Perception becomes Microsoft Perception and will be generally available, with new capabilities that expand what agentic defense can do across your environment. A cyber stack built for agentic security Defending with AI takes more than adding individual agents to existing tools. It requires a new cyber stack connecting signals, intelligence, and orchestration so agents can work together across the entire security lifecycle. Microsoft Perception turns signals from across the digital estate into actionable context for specialized models and teams of agents. Its orchestration harness assigns and coordinates work across investigation, protection, and remediation, while security teams set priorities and approve critical actions. Microsoft Perception is now generally available When we introduced Project Perception in July, we set out a vision for defense that could operate at the speed and scale of AI. Now generally available as Microsoft Perception, we’re advancing that vision with expanded agentic capabilities. Perception brings together specialized red, blue, and green team agents to transform how defenders protect their environments. Red agents find and validate potential paths to compromise. Blue agents triage alerts, investigate attacks, and turn threat intelligence into detections. Green agents identify the exposures that matter most and coordinate remediation to reduce risk. Together, they connect discovery, investigation, and action in a continuous cycle of defense—with humans in control. Playbooks orchestrate the right agents around a specific security outcome and deliver results defenders can review and act on. Teams can start with a playbook or describe what they need in chat, then follow the work and review results through an integrated session experience. Stop threats at machine speed Microsoft Perception is advancing the agentic SOC, helping security teams scale operations and focus their expertise where it matters most. At Microsoft Ignite, we’re introducing new capabilities that build on that progress. Scale triage and investigation Security teams need to move from an alert to a confident understanding of the attack without being constrained by queue size or analyst capacity. The Triage Agent can now triage about 80% of alerts in a typical Microsoft Defender queue*, reaching a verdict and explaining its reasoning so analysts can quickly review the evidence and act. For TÜV SÜD, the agent’s broader coverage helps analysts review each incident with instant, relevant context. Karsten Bode, Enterprise IT Security Architect at the company, explained: “The Triage Agent covers many incidents, performing analyses and making the right threats available to analysts. This gives us a helpful foundation for making faster, more effective decisions and the necessary next steps.” Once agentic triage narrows the focus to a much smaller set of confirmed threats that warrant deeper analysis, the Attack Investigation Agent can go deep on those incidents. It reasons across endpoints, identities, email, Active Directory, cloud resources, and relevant activity even when it didn’t generate an alert. The agent then reconstructs the attack timeline, identifies what was compromised, and recommends a prioritized response, such as containing an account or isolating a device. You can also invoke the triage and investigation agents automatically based on triggers that run their respective playbooks. This allows agents to begin work before an analyst opens an alert and handle multiple attacks in parallel. Teams can scale analysis as demand changes, while security practitioners stay focused on the evidence, decisions, and outcomes that need their expertise. Expand what your agents can see and act on Threats do not stop at platform boundaries. Microsoft Perception brings third-party signals and external intelligence into agentic workflows, giving agents broader context to understand what is happening and helping teams turn new insight into protection faster. The Attack Investigation, Detection Authoring, and Triage Agents can now reason across signals from specific third-party vendors, with support for additional vendors coming in the months ahead. The Attack Investigation Agent can follow evidence across more sources to build a more complete picture of an attack, the Detection Authoring Agent can use third-party data, tables, and schemas to build more tailored detection logic, and the Triage Agent can bring third-party alerts into its analysis to better assess and prioritize potential threats. Together, this helps agents make sense of more of the data already flowing through the SOC instead of stopping at the boundaries of the Microsoft security stack. Agent Supported third-party signals Attack Investigation Okta, AWS Detection Authoring Okta, Salesforce, AWS Triage Okta, Salesforce That same principle extends beyond telemetry to threat intelligence. Valuable insight into a new threat may come from Microsoft, but it may just as easily surface in another security vendor’s research or a government advisory. The Threat Intelligence Agent can now turn external threat articles into structured findings, extracting the attacker behaviors, techniques, and indicators that matter and making that understanding available for action. With the Detect emerging threats playbook, the Detection Authoring Agent can then translate that structured intelligence into tenant-aware KQL detection recommendations. Before a detection is activated, it is tested against historical tenant data to estimate alert volume and surface tuning needs, with analysts reviewing and adjusting the recommendation before deployment. Think like an attacker with advanced code scanning Security teams do not need another list of possible vulnerabilities. They need to know which weaknesses attackers can actually exploit and how to reduce the risk. Microsoft Perception is extending red-team testing to help defenders move from potential vulnerabilities to focused action. With the Run web pen test playbook, defenders can safely conduct a penetration test. The Web App Pen Test Agent tests an authorized web application from an attacker’s perspective. It examines areas such as authentication, APIs, injection and business logic, providing defenders with the evidence they need to confirm which weaknesses are actually exploitable. For each identified weakness, it provides evidence, affected endpoints, severity, and proofs of concept. In an independent benchmark conducted with Incalmo, the Web App Pen Test Agent achieved a 31.0% average task solve rate, about twice OpenAI Codex’s 15.2%, while staying within specified testing boundaries in 91% of safety-constrained tests. The results demonstrate its ability to help defenders focus on exploitable weaknesses, while operating within the intended scope of an authorized assessment. Read the full results here. The Protect web app playbook extends the workflow from finding exploitable vulnerabilities to helping protect against them. It first uses the Web App Pen Test Agent to find weaknesses. Then, the Risk Reduction Agent leverages those findings to recommend targeted remediations, such as Web Application Firewall (WAF) rule changes in Azure Front Door, that can reduce exposure while teams work toward permanent fixes. Both playbooks can optionally use code vulnerability scanning findings as input. One source for those findings is Codename MDASH, which can guide the agent toward relevant vulnerabilities to determine whether they are exploitable in the running application. When we introduced Codename MDASH earlier this year, we shared our approach to using a multi-model agentic system to uncover deep vulnerabilities in source code. At Ignite, MDASH can scan larger repositories more efficiently by breaking them into logical units and analyzing them in stages. Teams can track scan progress and spend less time on false positives. Expanded support for the MAI-Cyber model family—including MAI-Cyber-1.1-Flash and MAI-Cyber-2 in preview—further improves efficiency while maintaining the quality of security findings. Teams can run MDASH where development already happens—from the Microsoft Defender portal and command line to CI/CD workflows such as GitHub Actions and Azure DevOps. Developers can then review generated fixes locally or through an automated pull request. Shrink your attack surface Security teams can see more vulnerabilities than they can possibly address. The challenge is knowing where risk demands action, and getting there before an attacker does. At Ignite, Perception is bringing exposure management and agents together to make that possible. The AI Readiness Initiatives page organizes recommendations and agentic actions into focused, outcome-driven programs of work. Each initiative organizes the risks, recommendations, and agentic actions needed to achieve a specific security outcome, so teams can manage exposure as a coordinated effort rather than a series of disconnected fixes. One new initiative helps organizations manage the surge of vulnerabilities that arrives with each monthly Windows update. It brings the highest-priority findings into a recurring initiative, where agents can help teams prioritize what matters, plan the response, and track the work through completion. Instead of treating every vulnerability as equally urgent, teams can focus their resources on the exposures most likely to put critical assets at risk. The Reduce exposure risk playbook calls on the Risk Reduction Agent to assess the available paths for reducing risk and recommend the right response. When a permanent fix is ready, the agent can coordinate patching through Microsoft Intune, with IT maintaining control of deployment. When patching is not immediately possible or could disrupt the business, the agent can help shield vulnerabilities by providing targeted protection while teams work toward the permanent fix. Learn more about Microsoft Perception for exposure management and Microsoft Perception in Intune. Secure your AI and agents AI agents are gaining access to more sensitive data, tools, and systems—and their risk can change as their permissions, connections, and behavior evolve. Security teams need to identify the agents that require attention, understand why, and prioritize where to act. At Ignite, we’re extending agentic security to help teams investigate and reduce AI agent risk. Together, Microsoft Perception and Microsoft 365 E7 bring signals, context, and registry data into a consistent view of the agent landscape. This foundation helps teams investigate an agent’s access and behavior, understand where concerns may arise, and identify the changes that can reduce the most risk. Playbooks turn that understanding into prioritized remediation and coordinated response, helping teams contain risk and respond as agent behavior and access evolve. The result is a more adaptive approach to AI security, helping teams prioritize what matters most and respond at machine speed. Learn more in our recent blog. Put agentic security to work, and stay in control Agentic security should strengthen the workflows, integrations, and controls security teams already rely on. With APIs and Model Context Protocol (MCP) support, defenders can start Microsoft Perception playbooks from ITSM, SIEM, or SOAR tools, then bring agent sessions and artifacts back into those workflows. The Government of Nunavut has explored that approach. As Arshad Sheikh, Senior Security Officer, Government of Nunavut, put it: “Using the Attack Investigation Agent and Recon Agent through Perception MCP and now with Microsoft Foundry, gave us a practical view of how agentic security can support defenders with deeper context and more efficient investigations while keeping people at the center of the process.” And as organizations bring agents into more workflows, they can plan and scale that usage with a clearer view of cost. New usage dashboards and a cost calculator provide visibility into agent activity and spending, helping teams plan investment and connect what they spend on agentic security to the outcomes they achieve. See agentic security in action at Microsoft Ignite Microsoft Perception brings agentic security to life across the security lifecycle—helping teams stop threats, think like an attacker, minimize exposure risk and secure the AI estate. Experience these capabilities firsthand and meet the teams building them at these sessions: Project Perception Roadmap: New Capabilities, Use Cases & What's Next, with Hari Pulapaka on November 17 th at 2 pm PST. Find and fix code vulnerabilities faster with Codename MDASH, with John Nix and Henrique Oliveira on November 17 th at 4:30 pm PST. Project Perception: How Agentic Security Works, with Naama Ben Dov and Corina Feuerstein on November 19 th at 10:15 am PST. Not attending in person? Watch the sessions on demand and explore Microsoft Perception to see how AI-powered security can strengthen your organization’s defenses. Follow the Microsoft Perception Community for the latest updates. *Based on Microsoft’s analysis of approximately 8 million alerts generated across Microsoft Defender workloads (identity, endpoint, email and collaboration, cloud apps, and cloud security), covering the highest-prevalence SOC-actionable alerts of low-to-high severity. Approximately 80% of alert volume was determined to be within Triage Agent coverage. This percentage represents aggregate coverage across the analyzed population and is not a guaranteed triage rate for an individual customer.2.7KViews3likes0Comments