tech
Microsoft wants AI agents fixing bugs before hackers find them
The new tools come as cybersecurity and AI companies race to help defenders keep pace with attackers.
TL;DR
- Microsoft introduced Project Perception, a cyber-specific AI platform with AI-powered security agents.
- The agents are designed to help organizations identify, prioritize, and patch software vulnerabilities faster.
- Three initial agents (Red, Blue, Green) were unveiled to handle vulnerability identification, risk assessment, and patching.
- MAI-Cyber-1-Flash is Microsoft's first in-house trained security AI model, performing 95% of the work of their previous MDASH system.
- The combination of MAI-Cyber-1-Flash and GPT-5.4 achieved a 95.95% score on the CyberGym benchmark, surpassing competitors.
- Microsoft will make MAI-Cyber-1-Flash available via Azure AI Foundry, not publicly released.
- The strategy of using a smaller, specialized model for most tasks aims to reduce costs.
- This move follows similar AI cybersecurity announcements from Google and Cisco.
- Microsoft acknowledges that organizations will gradually build trust before granting full autonomy to AI security agents.
- Public preview of MAI-Cyber-1-Flash begins next week, with plans to expand Project Perception.