Microsoft launches Project Perception & AI-Cyber-1-Flash

Microsoft has launched Project Perception, an agentic security system for organisations, and introduced Microsoft AI-Cyber-1-Flash, its first AI model focused on cyber security.

The announcements mark a broader push by Microsoft to apply specialised AI models and autonomous agents to security work as companies face more complex digital estates and faster-moving attacks.

Project Perception is designed to continuously identify, assess and reduce security risk across an organisation's environment. It brings together security signals, contextual data, AI models and specialised agents in a structure intended to move security teams from reactive investigations to ongoing risk reduction.

The system is organised around three groups of agents. Red agents identify possible attack paths and software weaknesses before they are exploited. Blue agents investigate those findings and determine which issues amount to material risk. Green agents take corrective action.

Microsoft said the three groups form a closed-loop process that continuously discovers, evaluates and improves an organisation's security posture over time. The approach is designed to keep human operators in control while allowing software to handle parts of the analysis and response cycle at machine speed.

New model

Alongside the system, Microsoft unveiled Microsoft AI-Cyber-1-Flash, a cyber security model built for software vulnerability analysis. It forms part of what Microsoft described as a multi-model strategy that uses both frontier AI models and more narrowly specialised models for different security tasks.

The model is being introduced first in a software vulnerability management workflow inside MDASH, which Microsoft described as its software vulnerability multi-model team of agents. According to Microsoft, the combination of MDASH and Microsoft AI-Cyber-1-Flash scored 96% on the CyberGym benchmark, 12 points above Mythos, while cutting costs by almost 50% compared with the current MDASH configuration on the market.

Microsoft said no single model is best suited to every security task. Instead, Project Perception is built to select models based on a mix of quality, reliability, latency and cost. Microsoft argued that always-on security operations require tighter economic control as well as technical accuracy.

Cyber stack

The launch also reflects Microsoft's effort to define what it calls a new cyber stack for AI-era defence. In that structure, signals and sensors provide visibility across assets, identities, data, cloud systems, applications and AI systems, while security context turns those signals into a form agents can use more efficiently.

Above that, models provide reasoning, a coordination layer manages interactions between models and agents, and actuators translate decisions into operational actions. Microsoft said the combination is intended to create a continuously learning security system that can understand risk, adapt to changing conditions and improve outcomes over time.

A central element of that design is security context. Microsoft said it draws on visibility across identities, endpoints, applications, data, cloud environments and AI systems, alongside threat intelligence and operational security experience, to build a near real-time view of an organisation's assets, relationships, risks and activity.

This allows agents to work from a shared understanding of the environment rather than repeatedly pulling raw signals together from scratch. Microsoft said that reduces the time, computing demand and cost associated with reasoning over security problems at scale.

Growing pressure

The launch comes as large technology companies seek to position AI not only as a business productivity tool but also as a defence mechanism against attacks that can now be automated and scaled more easily. Microsoft framed the change as a shift in the economics of cyber security, with attackers able to generate exploits faster and run broader campaigns than in earlier periods.

Hayete Gallot, Executive Vice President, Microsoft Security, set out the company's position in a statement accompanying the launch.

"The physics of cybersecurity are changing. Autonomous systems can now reason, adapt and operate continuously. At the same time, the cost of offense is falling, while the volume, velocity and complexity of what must be secured continues to grow. Attackers can generate exploits faster, scale campaigns further and operate with unprecedented efficiency. The approaches built for a world of human actors cannot keep pace with a world of AI, agents and machine-speed attacks. Security needs a new Cyber Stack," said Hayete Gallot, Executive Vice President, Microsoft Security.

Gallot also described Project Perception as an early step in that transition.

"Security has always been a race between attackers and defenders. AI changes the speed, scale and economics of that race. Defenders need systems that can continuously perceive, reason and act alongside them. Project Perception is how we begin to build that future," said Gallot.