Microsoft Says New Cybersecurity AI Model Helps MDASH Score 95.95% at Half the Cost
Microsoft's multi-model vulnerability discovery and repair harness, MDASH, has introduced its first cybersecurity-specific model.
According to the firm, MDASH achieved 95.95% on CyberGym utilizing MAI-Cyber-1-Flash and GPT-5.4. Additionally, it states that the setup is 50% less expensive than its best MDASH combination of GPT-5.4, GPT-5.4 small, and GPT-5.3 Codex. Only authorized MDASH clients have access via an Azure AI Foundry private preview.
Up to 90% of MDASH tasks can be handled by MAI-Cyber-1-Flash, with GPT-5.4 reserved for the most difficult 10%. It is solely accessible within MDASH; it is neither a general-purpose application programming interface nor a stand-alone public model.
The headline score does not belong to the new model alone, but rather to MDASH using MAI-Cyb...