Microsoft unveils new cyber model, agentic security tools to fight hackers

Microsoft unveiled Project Perception, a platform of specialized security AI agents that identify vulnerabilities, assess risk and deploy patches, alongside MAI-Cyber-1-Flash, a purpose-built security model that scores 95.95% on exploit-generation benchmarks versus 83% for Anthropic and OpenAI competitors. The model will be available through Azure AI Foundry rather than open release, positioning Microsoft to capture security-specific AI workloads at enterprise scale.

This Wire brief sits within Fusion42's coverage of Cybersecurity. Wire is Fusion42's founder-focused intelligence feed: each story is connected to the funds and startups it names — every one with a live profile on Raise or Scout — so founders can follow the capital and the momentum behind the headline rather than just the headline itself. Wire analysis is one of the live surfaces Arthur reasons over.