Microsoft debuts specialized AI model and agentic security platform to rival Anthropic and OpenAI

Microsoft unveiled two new cybersecurity offerings this week: MAI-Cyber-1-Flash, a specialized AI model for vulnerability detection, and Perception, an autonomous platform that coordinates multi-agent teams to identify and fix security flaws at scale. The announcements signal Microsoft's intent to challenge established competitors like Anthropic, Google, and OpenAI in the rapidly expanding AI security space.

The models and platform#

MAI-Cyber-1-Flash is purpose-built to discover intricate vulnerabilities within large, complex code repositories. The model integrates with MDASH, Microsoft's framework for automated software vulnerability scanning and repair. According to the company's benchmarks, the model delivers superior performance compared to rival offerings while also reducing costs.

Perception operates as an orchestration platform that deploys coordinated teams of autonomous agents to handle different aspects of security operations. The system automates detection, prioritization, and remediation of bugs while also working alongside MDASH for broader security scanning. The platform can tackle the full security lifecycle—from identifying problems to deploying fixes—where human teams once required hours of manual effort across multiple specialties.

How the agent teams work#

Perception deploys three types of agent teams, each with a distinct function:

• Red teams simulate realistic attack scenarios, modeling tactics and likely exploit paths that threat actors might pursue
• Blue teams hunt for existing bugs and prioritize them based on severity and exploitability
• Green teams execute remediation and implement protective measures against identified vulnerabilities

Dave Weston, the lead engineer on the project, highlighted the efficiency gains: work that previously demanded hours from multiple specialized security experts—vulnerability researchers, remediation engineers, and others—can now be completed in minutes, complete with both discovery and code-level fixes.

The competitive landscape#

Microsoft's push reflects an industry-wide race to combine AI capabilities with cybersecurity. Anthropic released its Mythos security platform earlier this year through a limited partner program called Glasswing. launched its own security solution in May via an initiative called Day Break. The competition underscores both the demand for AI-driven defense tools and the urgency with which major AI labs are moving to capture this market.