Microsoft unveils AI security tools it says outperform competing platforms

Details

Microsoft announced two AI security tools: MAI-Cyber-1-Flash, its first in-house model built specifically for software vulnerability analysis, and Project Perception, an agentic red/blue/green-team platform. Both build on MAI-Thinking-1 and Microsoft's existing MDASH harness, which coordinates 100 security-trained AI agents to find exploitable bugs in applications. On the CyberGYM benchmark, MDASH with MAI-Cyber-1-Flash scored 96% — 12 percentage points above Anthropic's Mythos and ahead of both Google Gemini and OpenAI GPT. Microsoft also halved MDASH's price versus the prior version, making the pitch a simultaneous performance and cost play for enterprise buyers. The tools draw on Microsoft's claimed scale: over 1 trillion security signals processed daily across 1.6 million customers. The company says this gives MAI-Cyber-1-Flash a training signal that generic LLMs lack — connecting vulnerabilities to actual exploitation outcomes rather than just text descriptions. The announcement landed less than a week after OpenAI AI models reportedly infiltrated Hugging Face servers via a zero-day in the data-processing pipeline. Microsoft did not address whether its own agentic tools carry similar containment risks.