Security
27 July 2026
Microsoft Unveils AI Model for Vulnerability Detection
N
New-ZZZ desk
X @MicrosoftAI · 1 week ago
Microsoft AI has introduced MAI-Cyber-1-Flash, an internal model for finding and fixing vulnerabilities in large codebases. It operates as part of MDASH, a platform where multiple AI agents work together on cybersecurity tasks.
The combination of MAI-Cyber-1-Flash and GPT-5.4 outperformed Mythos by 12 points on the CyberGym benchmark. Microsoft also claims that this MDASH configuration costs 50% less than the previous setup based on GPT-5.4, GPT-5.4 mini, and GPT-5.3 Codex.
Why it matters
- —The new model automates vulnerability detection and remediation in large software projects.
- —The CyberGym result indicates that the multi-agent system outperforms the competing Mythos system.
- —A 50% cost reduction could make automated code analysis more accessible to large teams.
Key facts
- MAI-Cyber-1-Flash operates within the MDASH multi-agent platform.
- The platform is designed to detect and remediate vulnerabilities in large-scale codebases.
- The combination of MAI-Cyber-1-Flash and GPT-5.4 outperformed Mythos by 12 points in CyberGym.
- The new configuration is offered at 50% less than Microsoft's previous best option.
The full text is in the original source. Here we provide a brief summary and key facts.
