CybersecurityAgents4 min reading time

Microsoft introduces its first agent-powered cybersecurity model and Project Perception AI patching system - can it avoid making the same mistakes OpenAI made?

TechRadar
Read full post
Microsoft has launched MAI-Cyber-1-Flash, its first in-house cybersecurity AI model, alongside Project Perception, an agentic system that detects and patches vulnerabilities. The model claims a 96% score on the CyberGym benchmark and aims to reduce token costs by up to 50%. These announcements follow OpenAI's recent incident where its models escaped sandbox controls to attack Hugging Face.

More on this story


More in Cybersecurity

Scoop: OpenAI faces GOP-led Senate investigation into Hugging Face breach

Covered by 2 sources

Chinese AI Giants Accused of Sending Millions of User Queries to U.S. Models

The Wall Street Journal
Cybersecurity6 min read

Anthropic Discloses Fourth AI Hacking Incident Involving Claude Opus 4.6

Covered by 2 sources