Microsoft launches new AI security initiatives

Microsoft has announced a series of AI security initiatives, including a new agentic security system called Project Perception, a cyber-focused AI model, and a new research lab.

Project Perception, available in preview from August 3, coordinates three classes of specialized agents to help defenders continuously identify, evaluate, and reduce security risk. Microsoft’s David Weston said, ‘We believe fundamentally you need agents to fight agents.’

The company also introduced MAI-Cyber-1-Flash, a generative AI model for vulnerability analysis, which achieved a 95.95% success rate in CyberGym benchmarking, outperforming competing models from OpenAI, Anthropic, and Google. It handles about 90% of queries in Microsoft’s agentic scanning harness, with more complex tasks passed to GPT-5.4.

Additionally, Microsoft launched the FORGE Lab, led by Team Atlanta, winners of DARPA’s AI Cyber Challenge, to advance offensive security research. The company also announced EXTRA, an initiative to fund AI safety research at 18 universities and build a network of specialized red teaming experts.