Microsoft Unveils Cybersecurity Model and Platform to Tackle AI-Driven Threats

Key points

• Microsoft launched MAI-Cyber-1-Flash, its first cybersecurity model, at a San Francisco event on Monday.
• The Perception platform deploys AI agent teams to automate vulnerability detection and fixing.
• Both tools enter preview on November 3.
• Microsoft claims MAI-Cyber-1-Flash beats Gemini, Mythos 5 and rival GPT variants on the Cyber Gym benchmark.

Microsoft launched its first cybersecurity-dedicated model, MAI-Cyber-1-Flash, at a small San Francisco event, taking direct aim at Anthropic and Google. It's a bold move from a company that already credited AI with finding a record 570 security flaws in a single monthly patch cycle back in July.

The model is built to power MDASH, Microsoft's harness for identifying and fixing software vulnerabilities. Mustafa Suleyman, CEO of Microsoft AI and co-founder of DeepMind, said MAI-Cyber-1-Flash paired with GPT 5.4 inside MDASH outperforms Gemini, GPT 5.5 Cyber, GPT 5.6 Sol and Mythos 5 on Cyber Gym, which he called "the golden benchmark." "We're shipping this into production immediately," he added. Microsoft also says the model is more cost-effective than its rivals, though no pricing was disclosed.

How does the Perception platform work?

Perception uses three AI agent teams. Red teams simulate potential attacks and map likely threat actors. Blue teams detect and triage existing bugs. Green teams apply fixes. Dave Weston, Perception's lead engineer, put the efficiency claim plainly: "We've gone from this taking hours and hours of manual work from multiple specialized folks... And in minutes, we have a fix for all of this."

Hayete Gallot, Microsoft's vice president for security, framed the platform's purpose around the attacker-defender arms race: companies must now "defend against AI with AI at the scale and speed that the attackers have."

What does this mean for businesses?

Hackers are already using AI to run faster, more targeted attacks. Perception gives defenders a comparable tempo, automating work that previously required teams of application security specialists and remediation engineers. That's the real number to watch: not the benchmark win, but how much specialist labour this actually replaces in practice once the November 3 preview opens.

Anthropically launched Mythos through a limited partner program called Glasswing, and OpenAI launched its own security solution in May through a program called Day Break, so Microsoft is the last of the three big labs to show its hand here.

Common questions

What is MAI-Cyber-1-Flash?

MAI-Cyber-1-Flash is Microsoft's first AI model built specifically to find vulnerabilities in complex code, designed to run inside the MDASH vulnerability-management harness.

How does Perception help in cybersecurity?

Perception automates the full vulnerability workflow, from attack simulation through bug detection to patching, cutting a process that once took many hours of specialist work down to minutes.