Microsoft unveils AI security tools with benchmark claims and cost advantages

Microsoft announced two AI security tools—MAI-Cyber-1-Flash and Project Perception—both currently in preview. The company describes MAI-Cyber-1-Flash as a compact, code-heavy security model trained on internal Microsoft data from decades of vulnerability patching and security incident responses across its products. Microsoft says the model processes more than 1 trillion security signals daily and learns from 1.6 million customers.

MAI-Cyber-1-Flash is integrated into MDASH, a multi-model agentic scanning harness introduced in May. Microsoft reports that MDASH with MAI-Cyber-1-Flash achieved a 96% score on CyberGYM, a standard benchmark test. The company claims this score is 12 points higher than Anthropic’s Mythos and also exceeds Google Gemini and OpenAI GPT.

Microsoft states the new MDASH costs half as much to use as the previous version. The second tool, Project Perception, is a collection of specialized AI agents that perform red-, blue-, and green-team functions for finding vulnerabilities, assessing risk, and taking corrective actions. Microsoft says the platform selects models based on task effectiveness and cost, aiming to perform 90% of tasks at lower cost than competitors.

The announcements follow a reported security incident involving OpenAI models on Hugging Face, though Microsoft did not reference that event. The company did not detail safeguards to prevent its new tools from going rogue. Microsoft frames the tools as a response to an increase in AI-accelerated cyberattacks and the limitations of traditional security approaches.