Microsoft has launched an agentic AI system that uses teams of specialised agents to find and fix cyber risks.
Project Perception brings together AI agents that simulate attacks, investigate threats and address security weaknesses. The system will enter public preview on 3 August.
The platform divides these tasks among three groups. Red team agents identify possible attack paths before they are exploited, while blue team agents investigate activity and determine which risks require attention. Green team agents then take corrective action and strengthen defences.
Microsoft designed the agents to work together in a continuous cycle that tests, evaluates and improves an organisation’s security posture. Human security teams retain oversight of the system and its actions.
Hayete Gallot, Executive Vice President of Microsoft Security, said traditional security systems may struggle to keep pace as AI allows attackers to create exploits faster and automate larger campaigns.
Multi-Model Approach Targets Security Costs
Project Perception draws signals from identities, endpoints, applications, data, cloud infrastructure and AI systems. It uses this context to prioritise threats and coordinate responses across an organisation’s digital environment.
The system also uses several AI models for different security tasks rather than relying on one model. Microsoft expects this approach to balance performance and operating costs.
Its first use case focuses on software vulnerability management. Microsoft is integrating its MAI-Cyber-1-Flash model into MDASH, its team of agents for identifying and assessing software vulnerabilities.
According to Microsoft, the combined system achieved a 96% score on the CyberGym benchmark, 12 percentage points above Mythos. The company also reported cost savings of almost 50% compared with the current MDASH configuration.
Microsoft plans to use MAI-Cyber-1-Flash across more Project Perception security workflows after the initial vulnerability management rollout.
Featured image: Edited by Fintech News Singapore, based on image by wayhomestudio via Magnific
