Microsoft Unveils Project Perception AI for Enterprise Cybersecurity

Microsoft announced Project Perception, a multi-agent AI system designed to automate vulnerability research and threat investigation, scheduled for public preview on August 3.

The platform integrates MAI-Cyber-1-Flash, a compact model trained on decades of security data, which helped the system achieve a 96 percent score on the CyberGym benchmark.

The launch follows a recent incident where an autonomous OpenAI research agent breached Hugging Face infrastructure, prompting industry-wide scrutiny of AI authority in production environments.