Microsoft has unveiled a new cybersecurity-focused AI system called MDASH, claiming it outperformed rival models from Anthropic and OpenAI in tests designed to hunt down software vulnerabilities—and did so while dramatically cutting costs.
A New Contender in Automated Security
The company says MDASH is built to coordinate large numbers of AI agents working in parallel to probe software for weaknesses. According to Microsoft, the system can marshal more than 100 agents simultaneously, allowing it to scan codebases and identify flaws at scale rather than relying on a single model to do the heavy lifting.
The headline claim from Microsoft's testing is that MDASH bested both Anthropic's Claude Mythos and OpenAI's GPT-5.6 Sol in benchmark evaluations focused on finding security defects. In an industry where AI-assisted vulnerability discovery is quickly becoming a competitive battleground, edging out two of the biggest names in the field is a notable marketing win.
Microsoft says its new system can find software flaws at half the cost of its previous best-performing setup.
Cost Efficiency as the Real Selling Point
Beyond raw performance, Microsoft is emphasizing efficiency. The company says the new configuration can uncover software flaws at roughly half the cost of its current best-performing MDASH setup. For enterprises weighing the price of running fleets of AI agents against the value of catching bugs before attackers do, that economic angle may prove just as persuasive as benchmark rankings.
The push reflects a broader trend across the tech sector, where companies are racing to deploy AI agents capable of automating the tedious and expensive work of security auditing. Multiple AI agents operating in concert can theoretically cover far more ground than human researchers alone.
Key points from Microsoft's announcement include:
- MDASH orchestrates more than 100 AI agents at once to hunt vulnerabilities
- The system reportedly outperformed Claude Mythos and GPT-5.6 Sol in testing
- Microsoft claims the setup halves the cost compared to its prior best configuration
As with any vendor-run benchmark, the results come from Microsoft's own evaluations, and independent verification will be key to gauging how MDASH performs against competitors in real-world security scenarios.