Strengthening secure software at global scale: How MSRC is evolving with AI

Microsoft on April 7, 2026 outlined how the Microsoft Security Response Center, or MSRC, is adapting its vulnerability-discovery and response workflows around new AI capabilities. In the post, MSRC engineering head Tom Gallagher argues that cybersecurity is entering a phase where defensive work is less constrained by human time and scale, and Microsoft is trying to integrate that shift into both its internal processes and customer-facing platforms.

What Microsoft says has changed

Microsoft says it already processes thousands of vulnerability reports every year and coordinates fixes across its portfolio. The new development is that AI systems are becoming strong enough to contribute meaningfully to discovery and remediation. Microsoft says it evaluated an early snapshot of Claude Mythos Preview with CTI-REALM, its open-source benchmark for real-world detection engineering tasks, and saw substantial improvement relative to prior models.

According to the post, Microsoft believes recent models paired with advanced security tooling are approaching the level of experienced human security researchers for some discovery work. The company also stresses that AI systems can run continuously, which means they can examine broader attack surfaces and identify more issues earlier in the software lifecycle.

How MSRC plans to respond

MSRC says it is adding more automation to validate the quality and severity of findings and to support remediation at what it calls AI speed, while keeping human developers in the loop for correctness and quality. Microsoft also says it is using agentic red teaming and embedding these capabilities directly into software-development processes so issues can be identified and addressed as code is written and shipped.

Customer and ecosystem implications

The post also links Microsoft's internal work to external access. Microsoft says Azure customers participating in Project Glasswing will be able to use Claude Mythos Preview through Microsoft Foundry, subject to Anthropic's access terms. That makes the announcement more than a process update: it connects Microsoft's internal security operations, Anthropic's emerging security-focused model capability, and a path for customers to experiment with the same class of tools under managed platform controls.