Recent patches for Windows 11 could have been created by Microsoft's new Mythos rival

Microsoft is working on a product to find exploits in code using AI. It will work similarly to Anthropic Mythos but at a lower price point.

Microsoft is allegedly working on a next-generation security tool called Project Perception that will help organizations find and fix software vulnerabilities with the help of AI. Project Perception is expected to launch this month and will combine AI models from Microsoft, OpenAI, and Anthopic. This will give it greater coverage and help to keep operating costs down.

Perception will be able to use multiple models to scan for security holes by using a model router that will determine the correct AI model for the job before assigning the workload.

While Microsoft is trying to do what Anthropic’s Mythos model does, Mythos is very expensive to run. Microsoft wants to achieve the same goal but more efficiently. While Perception should be cheaper than Mythos, Microsoft has not finalized any pricing details yet.

Project Perception is the brainchild of Hayete Gallot, Microsoft’s head of security who took over the role in February. She has been reorganizing Microsoft’s security business around AI products while reducing investment in older offerings. Targeting AI products to businesses is probably the single-best way to monetize AI and claw back the massive spending that went into it. Ultimately, they have greater budgets to spend on AI compared to individual home consumers.

Microsoft is also a huge player in the field of cybersecurity so it doesn’t want to let Anthropic and OpenAI steal all the customers that could be got from offering vulnerability-hunting AI bots. The extra competition from Microsoft is also good for consumers because it will mean lower costs overall and firms are less likely to gatekeep features.

When Project Perception finally does launch, its availability could also be restricted, just like Anthropic’s Mythos and Fable models. Unlike your average LLM, these all have the ability of finding exploits in software that could be abused before patches get released, so they’re considered quite dangerous.

In the recent Patch Tuesday, Microsoft patched around 570 holes thanks to an AI tool it calls MDASH. It is unclear if Project Perception is related to this, it's quite possible.

Source: The Information via Firstpost

Source

Hope you enjoyed this news post. Feedback welcome.

Posted Saturday 18 July 2026 at 7:27 am AEST (my time).

News posts: 2023 5,800+ | 2024 5,700+ | 2025 5,700+ | 2026 (to end of June) 2,475

RIP Matrix