Microsoft launches Project Perception AI cybersecurity system with MAI-Cyber-1-Flash model, undercuts Anthropic's Mythos on cost
Microsoft $MSFT unveiled Project Perception, an AI security system built on its first in-house cyber model, MAI-Cyber-1-Flash. Microsoft says it beat Anthropic's Mythos 5 by 12 points on the CyberGym benchmark at half the cost. Public preview opens August 3, per TechCrunch.
Microsoft $MSFT unveiled Project Perception on Monday, an agentic AI security system that coordinates red, blue and green team AI agents to hunt, prioritize and fix software vulnerabilities inside customer networks. It runs on Microsoft's first in-house cybersecurity model, MAI-Cyber-1-Flash, built on the company's MAI-Thinking-1 reasoning model and reviewed by Microsoft's internal AI Red Team plus an outside assessor. Microsoft says the combination outperformed Anthropic's Mythos 5 by 12 points on the CyberGym benchmark while costing half as much to run, positioning it as a direct undercut of Anthropic's existing vulnerability-scanning tool, which several large enterprises already use. Public preview opens August 3, according to TechCrunch and Microsoft's own blog. The launch shows cybersecurity vendors now competing on AI model price as much as detection accuracy, a shift that could reshape how enterprises budget for security operations as they look to cut reliance on human analysts for routine vulnerability triage.