Microsoft Rolls Out New AI Security Initiatives
▼ Summary
– Microsoft launched Project Perception, an agentic security system using Red, Blue, and Green agents to identify vulnerabilities, assess risks, and strengthen defenses.
– The system coordinates three agent classes: Red agents find attack paths, Blue agents evaluate risk, and Green agents take corrective action.
– Microsoft announced MAI-Cyber-1-Flash, a generative AI model for cybersecurity, integrated into its security scanning harness.
– The MAI-Cyber-1-Flash model, enhanced by GPT-5.4, achieved a 95.95% success rate on the CyberGym benchmark, outperforming competitors like OpenAI and Google.
– Project Perception will be available in Preview mode for all Microsoft customers starting August 3.
Combating AI-powered threats demands AI-driven defenses. That was the core message from David Weston, Microsoft’s corporate vice president for AI security, as he addressed attendees at the company’s security launch preview on July 27.
During the event, the Redmond tech giant unveiled a wave of new AI and security products and initiatives. Chief among them is Project Perception, an agentic security system built to help cyber defenders continuously identify, evaluate, and reduce risk across their environments.
Microsoft’s Perception coordinates three distinct classes of specialized agents that collaborate to improve security posture over time. Red agents scan for potential attack paths and vulnerabilities before exploitation. Blue agents investigate findings, apply security context, and determine which threats represent meaningful risk. Green agents then take corrective action and strengthen defenses throughout the environment.
This framework mirrors Google’s AI Threat Defense platform, which was powered by Wiz’s Red, Blue, and Green agents when it launched in May 2026.
Speaking at the event, Hayete Gallot, executive vice president at Microsoft Security, emphasized the company’s unique visibility into global threats. “We sit at your identity, data, cloud, code and even AI level. If you add our security research, threat intelligence and red teaming efforts, you end up with even more signals,” she said. Microsoft processes roughly 100 trillion signals each day.
However, raw data alone isn’t enough. “If you were to apply an agent to that raw data, it would be very slow and you would get terrible results,” Gallot explained. “That’s why we are connecting and correlating all those signals so we can provide a ‘security context,’ which is organized efficiently for our agents.”
Weston added that Project Perception will be multi-model and demonstrated several “playbooks” based on real-world operations a security operations center (SOC) might face. “We believe fundamentally you need agents to fight agents,” he reiterated.
Project Perception will enter Preview mode for all Microsoft customers starting August 3.
In a parallel announcement, Gallot revealed MAI-Cyber-1-Flash, Microsoft’s first generative AI model built specifically for cybersecurity use cases, particularly software vulnerability analysis. Developed by Microsoft AI (MAI), the model is based on the company’s internal MAI-Thinking-1 reasoning model and has been integrated into the MDASH (multi-model agentic scanning harness).
Mustafa Suleyman, CEO of Microsoft AI, noted that the system is further enhanced by GPT-5.4 and has outperformed competing solutions from Anthropic, OpenAI, and Google in CyberGym benchmarking. The MAI-Cyber-1-Flash and GPT-4.5 enhancement achieved a 95.95% success rate on the CyberGym benchmark.
For comparison, OpenAI’s GPT-5.5 Cyber scored 85.6%, GPT-5.6 Sol reached 83.6%, Anthropic’s Mythos recorded 83.8%, and Google’s Gemini 3.5 Flash Cyber hit 83.2%.
(Source: Infosecurity Magazine)
