July 17, 2026

Microsoft’s MDASH AI beats Anthropic’s Mythos in bug detection

Microsoft has announced that its new AI system, MDASH, has surpassed Anthropic’s Mythos in effectiveness by achieving a score of 88.45% on the CyberGym security benchmark, compared to Mythos’s 83.1%. MDASH has already identified 16 new security flaws in Windows, contributing to Microsoft’s recent record of patching 570 security vulnerabilities in a single month—an increase of 316% from the previous year. This push to enhance security comes as Microsoft integrates more agentic AI systems across its operations in response to the evolving threat landscape, where faster exploit development by adversaries is prompting the company to recommend shorter deployment timelines for security updates.

MDASH: MDASH is Microsoft’s codename for a multi-model agentic security scanning system that orchestrates specialized AI agents to identify and validate software vulnerabilities. It was deployed internally to discover flaws in Windows components and integrated into broader security workflows.
CyberGym: CyberGym is a public benchmark developed by UC Berkeley researchers consisting of real-world vulnerability reproduction tasks drawn from open-source projects. It evaluates how effectively AI agents can find and reproduce software bugs across diverse codebases.
Anthropic: Anthropic is an AI company focused on building advanced language models with safety considerations. It developed the Mythos preview model, which has been evaluated on cybersecurity benchmarks for vulnerability discovery capabilities.
Microsoft: Microsoft is a leading technology company that develops software, cloud platforms, and AI solutions including security tools. Its MDASH system was recently introduced as a multi-model agentic AI harness for automated vulnerability detection in complex codebases such as Windows.

`json
{
“AI Integration”: “Microsoft is expanding the use of agentic AI systems like MDASH across its engineering and security operations to accelerate vulnerability identification.”,
“Benchmark Competition”: “Frontier AI models from multiple companies are being tested on shared cybersecurity benchmarks to compare their effectiveness in real-world vulnerability tasks.”
}
`

AI tools built by Emerald Force

Built and supported by Emerald Force.

You might also like