At a small event in San Francisco, Microsoft unveiled its first dedicated cybersecurity model, as well as a new AI-powered cybersecurity platform.
At a small event in San Francisco, Microsoft unveiled its first dedicated cybersecurity model, as well as a new AI-powered cybersecurity platform.
The company describes MAI-Cyber-1-Flash as a model built «to find complex vulnerabilities in large and complex codebases.» It is designed to run and support MDASH, Microsoft’s framework for detecting and fixing vulnerabilities in software, TechCrunch reports .
According to the company, MAI-Cyber-1-Flash is significantly more powerful (and cost-effective) than competing models when measured against a recognized cybersecurity AI benchmark.
«We are very, very excited to announce our results,» said Mustafa Suleiman, co-founder of DeepMind and current CEO of Microsoft AI. «We combined MAI-1 Cyber Flash with GPT 5.4 inside the MDASH environment — and this combination outperforms Gemini, GPT 5.5 Cyber, GPT 5.6 Sol, and Mythos 5 on Cyber Gym, which is our primary benchmark. It’s our gold standard. We’re launching it into production right away.»
The new security platform, called Perception, is designed to deploy teams of autonomous agents to assist and automate various cybersecurity processes, including error detection and remediation. The platform can also integrate with MDASH.
Noting that hackers are increasingly using artificial intelligence in their cyberattacks, Hayet Gallotte, Microsoft’s vice president of security, described Perception as a tool that will allow corporate defenders to «defend against AI with AI at the same scale and speed that attackers use.»
Perception uses agent-based «red teams,» «blue teams,» and «green teams.» Red teams can provide detailed simulations of potential attacks, describing the context around potential attackers and the likely vulnerabilities they could exploit. Blue teams specialize in identifying and initially assessing existing bugs, while green teams take «corrective action» to fix them.
Dave Weston, lead engineer at Perception, described the platform as a colossal step forward in improving efficiency for corporate defenders.
«Previously, this required many hours of manual work by many specialists across the security structure, from application vulnerability hunters to threat remediation engineers, and so on. Now, in minutes, we have a ready-made solution for all of this. We don’t just find and prioritize issues, we also provide detection, security improvements, and even ready-made code fixes.»
New security tools from Microsoft are going into public preview on August 3.
The market is already highly competitive among AI cybersecurity solutions. Earlier this year, Anthropic launched Mythos, a security platform that was made available to a select group of partner organizations through its Glasswing program. In May, OpenAI also launched its own cybersecurity solution through a program called Daybreak .
