Microsoft is preparing to launch Project Perception, a new AI-driven cybersecurity platform designed to spot and remediate software bugs at scale. The tool is expected to debut this month and is positioned by Redmond as a direct, lower-cost alternative to Anthropic's Mythos 5 security suite.
Multi-model routing keeps costs down
According to reporting from newsbytes and TechRepublic, Project Perception uses a smart routing layer that assigns each vulnerability-hunting task to the best-fit AI model - drawing from Microsoft's own MAI family alongside models from OpenAI and Anthropic. That heterogeneity is intended to preserve accuracy on hard cases while pushing routine triage to cheaper endpoints, a bill-of-materials play Anthropic cannot fully match with a single-model stack.
Enterprise distribution advantage
Project Perception rides on Microsoft's existing global distribution through Defender, Sentinel and Azure - a channel advantage that Anthropic, still primarily API-first, is racing to close. Redmond has also been retraining its enterprise sellers to differentiate against Anthropic's Mythos suite and OpenAI's own security offerings, according to a TechCrunch report last week.
Security AI arms race
The launch dovetails with a busy month for AI security. Anthropic recently extended free Fable 5 access for the third time in five weeks, while Microsoft's July Patch Tuesday shipped fixes for a record 570 flaws, including two exploited zero-days. That volume is precisely the workload Project Perception is being built to automate. See also our coverage of Defender's RoguePlanet zero-day and AnyDesk's CVE-2026-15682.
Reporting based on coverage from TechRepublic, newsbytes, Bloomberg and TechCrunch.