Benchmark Mythos! Microsoft is preparing an AI vulnerability detection tool that may be released this month
Microsoft is preparing a new artificial intelligence (AI) vulnerability detection tool, similar to the Mythos model owned by artificial intelligence company Anthropic, to meet the growing needs of enterprises for cybersecurity defense spending. According to a person familiar with the matter, the product, called ProductPerception, will be officially released within this month. It will combine Anthropic, OpenAI and a variety of artificial intelligence models developed by Microsoft itself to detect software vulnerabilities like Mythos and automatically repair these security flaws.
Microsoft is preparing a new artificial intelligence (AI) vulnerability detection tool, similar to the Mythos model owned by artificial intelligence company Anthropic, to meet the growing needs of enterprises for cybersecurity defense spending. According to a person familiar with the matter, the product, called Product Perception, will be officially released within this month. It will combine Anthropic, OpenAI and a variety of artificial intelligence models developed by Microsoft itself to detect software vulnerabilities like Mythos and automatically repair these security flaws. Although the exact price of Project Perception has not been finalized, people familiar with the matter said that Microsoft hopes to provide consumers with a more cost-effective alternative to Mythos, which is quite expensive. In addition, Microsoft plans to use a so-called "model router" to switch between OpenAI, Anthropic and Microsoft's own models based on the type of task, thereby keeping costs at a lower level. Project Perception is an initiative driven by Microsoft's new security chief Hayete Gallot to reinvigorate its security business sales and make it more competitive against rivals such as Anthropic and OpenAI. As the world's largest seller of enterprise office software, Gallot took over Microsoft's security department in February this year. It has recently conducted a comprehensive reorganization of the department, focusing on artificial intelligence tools while reducing investment in old products. MDASH beats Mythos in vulnerability detection It is worth mentioning that in May, Microsoft also launched an artificial intelligence network security tool called MDASH. MDASH is the abbreviation for "Multi-model Agentic Scanning Harness". It is a system of more than 100 specialized artificial intelligence agents working together through multiple models to discover, verify, and help remediate security vulnerabilities. It is reported that MDASH has discovered 16 previously unknown Windows vulnerabilities, including 4 serious remote code execution flaws, and assisted in patching 570 security issues within a month. All 16 newly discovered vulnerabilities were fixed in the May 2026 patch update, which means MDASH not only discovered the vulnerabilities, but discovered them quickly enough to be included in Microsoft's existing monthly security update cadence. Additionally, MDASH achieved a score of 88.45% on the CyberGym Detection Benchmark, the industry-standard AI-based vulnerability detection assessment. Anthropic's Claude Mythos preview scored 83.1%, while OpenAI's GPT-5.5 scored 81.8%. (