Microsoft Just Turned Its AI Loose on Hackers

Microsoft built a cybersecurity model and gave it a job security teams have wanted automated for years: find the hole before the bad guys do, then go fix it. The system is called Project Perception, and it runs three kinds of AI agents in a loop. One hunts for weaknesses the way an attacker would. Another decides which of those weaknesses actually pose real risk. A third writes and ships the fix.

The number worth sitting with is 95.95 percent. That’s how well Microsoft’s new MAI-Cyber-1-Flash model, paired with GPT-5.4, scored on a benchmark that tests whether a model can build a working exploit for a known vulnerability. Anthropic’s Mythos and OpenAI’s own cyber model both landed closer to 83 percent on that same test. Microsoft built something specifically for this fight, and it shows in the numbers.

David Weston, who leads security engineering at Microsoft, said work that used to take teams of specialists hours of manual digging can now get a fix in minutes. That’s a big claim, but security teams have genuinely been drowning for years, patching an endless stream of flaws while attackers use the same kind of AI tools to find new ones faster than humans can keep up.

There’s an obvious tension sitting underneath all of this. If AI can hunt bugs at machine speed, so can whoever is trying to break in. Microsoft is betting the defenders get there first, and it’s putting real weight behind that bet, with a public preview landing August 3.

Whether this actually tips the balance toward defense is going to take time to prove. But for once, the automation is being aimed at the people trying to keep the lights on, instead of the people trying to sell you something.

——

Follow: @Ali Demi
Book your free AI clarity call, NOW!
https://buff.ly/TpWy277

——

Sources:
https://blogs.microsoft.com/blog/2026/07/27/rethinking-security-for-the-age-of-ai/
https://techcrunch.com/2026/07/27/microsoft-launches-its-first-cyber-model-and-a-new-agentic-cybersecurity-system/
https://www.axios.com/2026/07/27/microsoft-unveils-new-cyber-model-agentic-security-tools-to-fight-hackers

Repost this. Thanks.