Project Perception is scheduled to enter public preview on August 3. The platform is designed to address evolving security challenges in the age of artificial intelligence by integrating advanced machine learning models into cyber defense workflows.
The system utilizes a multi-model architecture that combines frontier models with specialized cyber models. This approach aims to provide robust coordination through three specific classes of agents: red team, blue team, and green team agents.
Score achieved by MDASH with MAI-Cyber-1-Flash on CyberGym, an industry-leading benchmark.
Enhanced Performance and Cost Efficiency
The MDASH configuration featuring MAI-Cyber-1-Flash outperforms the Mythos model by 12 percentage points on the CyberGym benchmark. Furthermore, this specific configuration offers approximately 50% cost savings compared to current MDASH configurations.
Performance score of the MAI-Cyber-1-Flash integrated configuration.
Cost savings achieved by the MAI-Cyber-1-Flash configuration over current MDASH setups.
Integration and Vulnerability Management
As part of its integration into MDASH, the first scenario for MAI-Cyber-1-Flash focuses on software vulnerability management.