Professor Kim Leads New Cybersecurity Initiative at Microsoft
Microsoft has unveiled Project Perception , a new enterprise AI service that helps cybersecurity teams continuously assess and improve their defense posture. The service will be available in public preview starting August 3 and is based on a multi-model architecture , where the system selects the most appropriate AI model for each task, achieving a better balance between performance and cost.
The company also announced MAI-Cyber-1-Flash, a new AI model specifically designed to find vulnerabilities in complex codebases. Combined with GPT-5.4 through MDASH, Microsoft’s internal vulnerability research system, it achieved a 96% score on the CyberGym benchmark, outperforming Anthropic Mythos 5 and OpenAI GPT-5.6-Sol, at nearly half the computational cost.
MDASH was developed by the new FORGE (Frontier Offensive Research & Generative Exploration), led by Georgia Techprofessor Taesoo Kim. The group includes former members of Team Atlanta, which won the DARPA AI Cyber Challenge (AIxCC) to develop autonomous AI systems that detect and fix vulnerabilities in critical open source software.
The company also announced MAI-Cyber-1-Flash, a new AI model specifically designed to find vulnerabilities in complex codebases. Combined with GPT-5.4 through MDASH, Microsoft’s internal vulnerability research system, it achieved a 96% score on the CyberGym benchmark, outperforming Anthropic Mythos 5 and OpenAI GPT-5.6-Sol, at nearly half the computational cost.
MDASH was developed by the new FORGE (Frontier Offensive Research & Generative Exploration), led by Georgia Techprofessor Taesoo Kim. The group includes former members of Team Atlanta, which won the DARPA AI Cyber Challenge (AIxCC) to develop autonomous AI systems that detect and fix vulnerabilities in critical open source software.