Microsoft Unveils AI-Powered Cybersecurity Model with 50% Cost Savings

Microsoft has unveiled its latest AI-powered cybersecurity model, MAI-Cyber-1-Flash, designed to combat real-time threats. The model is optimized for complex codebases and can perform automated vulnerability identification and remediation. By utilizing MAI-Cyber-1-Flash for up to 90% of routine security tasks, Microsoft's unified system slashes token costs by 50%.

The release works alongside MDASH and Perception, a broad agentic system built to continuously monitor, patch, and close emerging security threat vectors. According to Microsoft executives Mustafa Suleyman and Hayete Gallot, the surge in AI capabilities has lowered the barrier for malicious actors seeking to probe massive software repositories for single vulnerabilities.

MAI-Cyber-1-Flash focuses heavily on code analysis and was built to autonomously process approximately 90% of standard security tasks. The model achieved a 96% success rate during benchmark testing on CyberGym, outperforming systems reliant on models like Mythos, Gemini, and GPT.

Alongside the new cyber model, Microsoft introduced Perception, an agentic security architecture featuring over 100 specialized agents that handle various defensive workflows. Microsoft plans to extend MAI-Cyber-1-Flash capabilities across broader security tasks beyond software vulnerability.