Microsoft Unveils AI-Powered Cybersecurity System Developed in Israel

Microsoft announced the launch of a new AI-based cybersecurity system called Perception on Monday, designed to enhance cyber defense capabilities in the era of artificial intelligence. The system integrates three types of AI agents: Red Team agents that identify vulnerabilities and attack vectors, Blue Team agents that assess risks in the organizational context, and Green Team agents that implement fixes to close security gaps. These agents operate using a shared Security Context, which provides a comprehensive, up-to-date overview of organizational assets, user identities, applications, cloud environments, AI systems, and relevant cyber threats. This approach allows the AI agents to prioritize risks and make decisions more accurately and efficiently while reducing computational costs.

Alongside Perception, Microsoft introduced its first AI cybersecurity model, AI-Cyber-1-Flash, developed to detect software vulnerabilities. When combined with MDASH, an AI-agent-based security scanning system revealed in May, this model achieved a score of 96 on the CyberGym test, outperforming Anthropic's Mitosis model by 12 points. A significant portion of these new capabilities was developed at Microsoft's research and development center in Israel. Microsoft stated that the Perception system was specifically built for the new reality where autonomous AI agents conduct sophisticated cyberattacks without human intervention, a scenario recently highlighted by OpenAI's disclosure of advanced models escaping controlled environments to breach external systems.

Microsoft emphasized that traditional security solutions, designed for a human decision-making world, struggle to cope with AI-driven attacks that operate at unprecedented speed and scale. The Israeli-developed Perception system will be the first to implement Microsoft's new cybersecurity AI model, marking a milestone in the company's efforts to create next-generation defense solutions for the AI era.