Microsoft has always been regarded as an embarrassing role in the AI competition with “no one has a better model, but has the most channels”, but this time they found a very beautiful corner. according to”The Information》Reporter Aaron Holmes exclusively reported that Microsoft is preparing an internal codename called Project Perception Anthropic’s AI security product is an attempt to challenge Anthropic’s prohibitively expensive Mythos vulnerability scanning system. This product uses a multi-model architecture, directly pulling in Anthropic Claude, OpenAI GPT and Microsoft’s own Phi/MAI, allocating it to different models according to different scanning tasks, strategically “betting on all three”, and using the most suitable model to run the most suitable job.
It’s not about being the strongest, it’s about planning the smartest
Anthropic’s Mythos is currently recognized as the strongest defensive security AI. The problem is that Anthropic controls the sales pipeline of Mythos very tightly. Only partners that have passed the Project Glasswing review can be authorized, and the total number is only about 40.
The price of Mythos is also quite scary. Data shows that the API fee of Mythos is 100% more expensive than Claude Opus and 82% more expensive than GPT. Microsoft’s strategy is simple: If you can’t make a Mythos-level model, it doesn’t matter. I’ll bring in Mythos technology, add GPT and my own model, and put it together into a cheap solution to sell to you. Moreover, this product is directly built into Azure Defender and Sentinel. Enterprises do not need to re-purchase or change platforms. They can use it by opening a license (somewhat similar to Japan Sakana Fugu practices).
More than just a Mythos replacement
In addition to Project Perception, Microsoft’s recent actions in AI security are actually more than this. Microsoft also has another internal AI tool called MDASH, which fixed 570 security vulnerabilities in one go during the last Patch Tuesday. Mustafa Suleyman, Microsoft’s head of AI, has publicly stated that the goal is for Microsoft to launch a truly cutting-edge model next year and no longer rely on distilling other people’s technology to catch up. .
I don’t know who will win yet, but there are more choices.
For enterprises, the biggest significance of the emergence of Project Perception is that there are now options for AI vulnerability scanning, and there is no need to queue up for Mythos. For Anthropic, this is a double-edged sword: the Claude model is the user, but Mythos, the most valuable exclusive selling point, is being turned into a “may not be exclusive” combination component.
The official launch date of Project Perception has not yet been announced, but sources estimate it will be launched this month. Global enterprise security spending is estimated to exceed $200 billion in 2026, and attackers are getting their hands on AI tools faster than defenders. In this situation where “time waits for no one”, having a choice is much better than having no choice.
Source: KOCPC Chinese