Microsoft Launches Agentic Security System with Red, Blue, and Green AI Teams
Microsoft has developed an innovative security system called Project Perception, which enters public preview on August 3. This system employs three distinct AI teams: red for vulnerability discovery, blue for risk assessment, and green for defense mitigation.
Project Perception:
- Objective: Coordinates AI agents to find vulnerabilities, assess risks, and fortify defenses at an unparalleled speed.
- Achievements: Microsoft’s custom cyber model, MAI-Cyber-1-Flash, achieves a remarkable 96% success rate on CyberGym, outperforming Anthropic’s Mythos by 12 points.
- Cost Efficiency: The system is claimed to deliver nearly 50% cost savings compared to existing setups by assigning the right models to specific tasks.
Key Features:
- Multi-Model Architecture: Utilizes frontier models for complex reasoning and specialized cyber models for high-volume, low-latency tasks.
- Comprehensive Visibility: Leverages Microsoft’s vast data from identities, endpoints, applications, data, clouds, and AI systems to take proactive measures.
- Continuous Operation: Unlike monthly patch cycles, the agents operate continuously, ensuring continuous security monitoring.
Competition and Impact:
Microsoft emphasizes its competitive edge over Anthropic’s Mythos, partly due to access to decades of enterprise defense data that Mythos lacks. The company aims for Project Perception to become the primary platform for AI-driven cyber defense, as demonstrated by its recent win in a government-led competition, Gold Eagle.
The public preview will allow enterprises to test the system’s effectiveness against real-world scenarios, marking a significant step forward in the evolution of cybersecurity.