Microsoft Project Perception: multi-model security routing cuts Anthropic Mythos cost by 50%

Microsoft is launching Project Perception by the end of July—an AI-powered vulnerability discovery platform that routes security analysis tasks across models from Microsoft, OpenAI, and Anthropic. The core innovation: intelligent model routing that reserves expensive frontier models (like Anthropic's Claude or OpenAI's GPT) for complex reasoning tasks and assigns cheaper, lighter models to routine work (log parsing, inventory checks, basic triage). This architectural choice cuts operating costs an estimated 50% below Anthropic's Claude Mythos, which carries API costs roughly double Claude Opus and 82% higher than GPT.

Project Perception is the commercial face of Microsoft's MDASH (Multi-Model Agentic Scanning Harness) internal security system, which discovered 16 Windows vulnerabilities in May 2026—4 of them critical RCEs—that were patched in that month's Patch Tuesday. MDASH uses 100+ specialized AI agents and achieved a 96.55% score on the CyberGym benchmark. The platform represents a major strategic shift under Microsoft security chief Hayete Gallot, who reorganized the division around AI products after arriving from Google Cloud in February. Microsoft's July 2026 Patch Tuesday, powered partly by MDASH, fixed 570 CVEs—a jump from traditional monthly volumes as AI-assisted discovery scales.

The competitive angle is direct: Anthropic launched Claude Mythos Preview in April but deliberately restricted it to ~50 approved Enterprise customers via Project Glasswing, citing security risks (the model can autonomously discover and weaponize zero-day exploits). Project Perception's multi-model approach sidesteps this bottleneck by distributing the workload. However, Microsoft has not yet disclosed pricing, eligibility criteria, or how widely available it will be; access may remain restricted to vetted enterprises and security researchers given the sensitive nature.

For enterprise security architects, this is a watershed moment: multi-model orchestration is becoming the default architecture for AI security tools, not a niche optimization. The cost advantage and access model make Mythos's exclusivity less sustainable. Microsoft's position (deep distribution through Azure, Defender, 365) means Project Perception could reshape vulnerability-discovery workflows at scale—but only if the company delivers on proof-of-concept credibility and maintains output quality.