Microsoft has quietly deployed one of its most ambitious security tools to Azure Government, and it changes the math on how government agencies can protect software that matters. Codename MDASH, a multi-model agentic code scanning system, is now available to select US government customers and authorized partners, bringing AI-powered vulnerability detection into the FedRAMP High-authorized cloud boundary where sensitive government code lives.
Why government software security looks different
The challenge facing government agencies isn't finding obvious vulnerabilities. Those get caught early. The problem is the subtle, deeply embedded weaknesses that only surface when you trace how data moves through an entire application stack. These are exactly the flaws well-resourced attackers hunt for, and they represent a small but dangerous category of risk that traditional scanning tools handle poorly.
Conventional security scanners work by pattern matching, scanning code against a catalog of known weaknesses. This approach is fast but degrades as the pattern list grows into the thousands. More importantly, it misses the kinds of vulnerabilities that require understanding how a program actually works. Security teams end up drowning in false positives, spending more time filtering noise than fixing real issues.
How MDASH actually works
MDASH takes a fundamentally different approach. Instead of pattern matching, it uses more than 100 specialized AI agents, each trained to recognize a specific category of weakness, to read and reason about source code the way a human security researcher would. The agents trace how information flows through a program, determining whether a potential weakness could actually be exploited by an attacker.
What makes this system distinctive is its debate architecture. A first group of agents identifies potential vulnerabilities. A second group then argues the case for and against whether each suspected flaw is genuinely reachable and dangerous. The system merges and deduplicates findings across the full scan, and where possible, it demonstrates the flaw rather than merely asserting it. What reaches the security team is a refined, prioritized list of actionable vulnerabilities, not raw noise.
On the public CyberGym benchmark of real-world vulnerabilities, MDASH scored 96.55. Microsoft reports similar results on first-party code validated against its historical vulnerability cases, suggesting the system performs consistently across government-style codebases.
The multi-model advantage
The architectural choice to use multiple AI models rather than a single system has practical implications. When two independent models disagree about a finding, that disagreement itself carries information. When one model flags a problem and another cannot argue it away, confidence in the finding increases. When two models independently agree, the signal is stronger still.
More importantly for long-term deployment, this design insulates the tool from the rapid pace of AI model iteration. A security tool built on a single model is only as good as that model's latest release. MDASH's harness directs whichever models are best for a specific task, so as newer, better models arrive, they can be adopted without rebuilding the system or losing previous work. The models change; the harness stays the same; the customer experience remains consistent within existing Microsoft security products.
Cost also matters at scale. Microsoft's own MAI model family is designed to make deep code scanning affordable across large codebases, and the newest addition to that family is expected to cut individual scan costs roughly in half. With more code in the world than can currently be reviewed at this depth, agencies are prioritizing software by mission importance and working through the list. Every cost reduction moves the line closer to full coverage.
Azure Government as the security boundary
Source code behind mission-critical systems is itself a security asset that requires additional protection, especially when the analysis process reveals vulnerabilities. MDASH is available in Azure Government as a feature of Microsoft Defender, using models available within the FedRAMP High-authorized Microsoft Foundry service. This keeps an agency's source code and all analysis results within an approved boundary.
Azure Government provides the isolated cloud environment these scans require: operated by screened US persons, built for federal, national security, and state and local government compliance obligations, including FedRAMP High authorization and Department of War accreditation for controlled workloads.
What happens when offense and defense use the same tools
There is an uncomfortable reality underlying this announcement. An AI system capable of finding vulnerabilities in a defender's software is equally capable of helping an attacker find ones to exploit. Threat actors are investing in AI capabilities, and the only durable advantage available to defenders is time: the gap between when a weakness can be found and patched, and before someone else finds it.
Microsoft has been running MDASH on its own software for months. The US government is not far behind, with customers across national security and civilian agencies already exploring its capabilities. For organizations that own software missions depend on, the question is no longer whether to examine code with a system like this. It is how quickly you can work through the backlog.