The honest view from inside technical leadership
AI News

AI News

Daily AI news with an engineering leadership take, each story credited to its original AI News report.

Security · AI News

Microsoft Perception Goes GA, and Its Agent Discovery Now Counts Your Engineers’ Coding Agents

Microsoft’s Ignite security launch is mostly about agents defending the network. The part engineering leaders should read first is about the agents already running on developer laptops.

Line-art illustration of a laptop with terminal windows linked to an identity badge, representing Microsoft Perception agent discovery of coding agents

Microsoft Perception, the company’s set of coordinated security agents inside Microsoft Defender, became generally available at Microsoft Ignite on November 17, 2026. In the same announcement, Microsoft said its agent discovery preview now recognises 60 types of local AI agents, including Claude Code, OpenAI Codex and GitHub Copilot, and gives each one it finds an identity the security team can govern.

Microsoft Perception Is GA, but Several of Its Agents Are Still in Preview

Aleš Holeček, president and chief architect of Microsoft Security, set out the release on the Microsoft Security blog. Perception links signals, context about the customer’s environment, and agents that investigate or respond, coordinated by playbooks. Microsoft says its own and partner sources produce more than 165 trillion signals a day through more than 500 connectors. The generally available pieces are Perception itself, core ISOC capabilities in Defender, and named agents for alert triage, attack investigation, detection authoring, threat intelligence, reconnaissance and posture prioritisation. The Risk Reduction Agent for cross-cloud remediation, the Web App Pen Test Agent, the code scanner known as Codename MDASH, temporary vulnerability shielding, and playbook integrations through MCP and APIs, including signals from AWS, Okta and Salesforce, are all in preview.

Humans Approve High-Impact Actions Under Criteria the Customer Writes

Microsoft’s stated model is bounded automation. A playbook fires only when an incident meets criteria the customer defines, and the blog says the customer’s team “signs off on high-impact actions.” Perception’s controls include containing devices, revoking identity access and blocking agents at the authorisation layer. AI News points out that Microsoft’s claims about the precision and token efficiency of its context layer are vendor statements with no independent results behind them yet. What counts as high impact is something each security team will have to write down, and that definition is where the real delegation decision sits.

Agent Discovery Turns Coding Agents Into Governed Identities

Microsoft says discovery added 35 local agent types, reaching 60, and also covers third-party cloud agents on AWS and Google Cloud. Discovered agents receive a Microsoft Entra Agent ID, which makes them subject to Conditional Access and identity governance. Intune policies can contain supported local agents on a device, while Defender and Purview data loss prevention watch what agents access at runtime. A preview feature called Dynamic Agent Risk scores each agent on exposure, blast radius and runtime behaviour. For an engineering organisation, this means the terminal agent a developer installed last month can now appear in the security team’s inventory as a named identity with a risk score attached.

Why This Matters for Engineering Leadership

Once coding agents are identities in Entra, the policy on how they are used will be written by someone, and engineering should help write it. If you run on Microsoft 365 and Defender, expect your security team to switch on discovery and see every Claude Code, Codex and Copilot instance across your fleet. Get ahead of that. Publish your own register of approved coding agents, the repositories and secrets each may touch, and the model providers they call. Agree with security which actions count as high impact for an agent on a developer machine, such as pushing to protected branches, reading production credentials or calling external networks, so Conditional Access and Intune containment rules match how your teams actually work. Ask for a test group before any containment policy reaches every laptop, since a blocked agent mid-incident is an outage for the developer using it. And treat the preview label seriously: discovery, Dynamic Agent Risk and the Risk Reduction Agent are not yet GA, so do not build hard release gates on them until Microsoft says they are.

FAQ

What is Microsoft Perception?

Microsoft Perception is a set of specialised Microsoft security agents plus the playbooks that coordinate them, delivered through Microsoft Defender. Microsoft announced its general availability at Ignite on November 17, 2026.

Which parts of Microsoft Perception are generally available?

Microsoft lists Perception, core ISOC capabilities in Defender, and agents including Alert Triage, Attack Investigation, Detection Authoring, Threat Intelligence, Recon and Posture Prioritization as GA. The Risk Reduction Agent, Web App Pen Test Agent, MDASH, vulnerability shielding and agent discovery remain in preview.

Does Microsoft Perception act without human approval?

Microsoft says customers set the criteria that trigger playbooks and sign off on high-impact actions. Lower-impact steps inside those limits can run automatically.

Can Microsoft detect Claude Code and Codex on developer machines?

Yes, in preview. Microsoft says discovery now covers 60 local agent types, including Claude Code, OpenAI Codex and GitHub Copilot, and assigns discovered agents a Microsoft Entra Agent ID.

What controls apply to a discovered coding agent?

Microsoft says discovered agents can be governed with Conditional Access, contained on the device with Intune policies, and monitored by Defender and Purview DLP at runtime.

How much data does Microsoft Perception use?

Microsoft says its own and partner sources produce more than 165 trillion signals a day, connected through more than 500 connectors.

Originally reported by Ryan Daws for AI News, November 18, 2026: Microsoft Perception coordinates security response agents. Primary source: Microsoft Security Blog, Aleš Holeček, “Security for the agentic era, powered by Microsoft Perception” (November 17, 2026).

Read more AI News