Microsoft launched MAI-Cyber-1-Flash in San Francisco on July 27, describing it as its first cybersecurity model, alongside Perception, an AI security platform that uses software agents to identify and remediate vulnerabilities.

The company said both tools will enter preview on Nov. 3.

Microsoft said MAI-Cyber-1-Flash works with MDASH, a multi-model agentic scanning harness already used by its security engineering teams and a small group of customers. Findings from MDASH flow into Microsoft Defender workflows, GitHub, and Azure DevOps pipelines.

The company has previously said MDASH scored 88.45% on CyberGym, which Berkeley Research Development Institute says measures whether AI agents can reproduce and verify 1,507 real-world vulnerabilities across 188 open-source projects.

Perception uses red-team agents to simulate attacks, blue-team agents to detect and triage bugs, and green-team agents to take corrective actions.

However, industry guidance says automated remediation still needs review in more complex cases.

The launch follows OpenAI’s Daybreak and Anthropic’s Mythos program, adding Microsoft to the growing market for AI-driven vulnerability discovery and defense.

🔗 Source: TechCrunch