Microsoft and Palo Alto Networks Use AI to Discover Dozens of Vulnerabilities in Their Own Code
Microsoft and Palo Alto Networks have separately reported that AI-driven vulnerability discovery systems found dozens of flaws in their own software, including critical remote code execution bugs in Windows.
Microsoft and Palo Alto Networks have separately reported this week that they have seen significant results after turning AI on their own code to find vulnerabilities. Advanced AI models such as Claude Mythos have sparked debate in the cybersecurity industry about what the vulnerability discovery landscape will look like going forward. While some organizations have confirmed that these AI models are a game-changer, others are skeptical of their actual performance.
Microsoft said on Tuesday that more than a dozen of the 137 vulnerabilities fixed with its latest Patch Tuesday updates were found by a new AI system called MDASH (multi-model agentic scanning harness) built by its Autonomous Code Security team. MDASH orchestrates more than 100 specialized AI agents across multiple frontier and distilled AI models, running a structured pipeline that moves findings through preparation, scanning, validation, deduplication, and proof construction. Different agents handle different roles: some identify candidate vulnerabilities, others argue for or against their exploitability, and a final stage attempts to construct inputs that actually trigger the bug. This multi-stage debate architecture means that a finding must withstand scrutiny before it reaches a human engineer.
According to Microsoft, MDASH was used to discover 16 of the vulnerabilities fixed with the latest Patch Tuesday updates. Four of them were rated critical, including unauthenticated remote code execution flaws in components such as the Windows kernel TCP/IP stack and the IKEv2 service. Microsoft also tested MDASH against pre-patch snapshots of two heavily audited Windows components, and the AI recovered 96% and 100% of the confirmed vulnerabilities found over the past five years. In addition, on the public CyberGym benchmark (which includes 1,507 real-world vulnerability tasks), the AI system achieved an 88% rating. MDASH is currently in limited private preview, with Microsoft inviting security teams to apply for early access.
Palo Alto Networks typically publishes 5-10 advisories per month. However, on Wednesday it published 26 new advisories, a record credited to its early access to frontier AI models such as Mythos. The company used AI to analyze more than 130 products across SaaS-delivered and customer-operated environments, including products obtained via the recent acquisitions of CyberArk, Chronosphere, and Koi. The 26 new advisories cover 75 vulnerabilities. While some were attributed to external researchers, the majority were detected internally using AI. The cybersecurity giant pointed out that none of the 75 vulnerabilities are critical and there is no indication that they have been exploited in the wild. Three high-severity vulnerabilities were detected, but their exploitation requires highly specific configurations to be weaponized.
Palo Alto Networks said it anticipates a surge in vulnerability discovery and patching as AI scanning becomes more widespread. The security firm believes organizations should act with urgency, as they have only a 3-5-month window to outpace adversaries. The company noted that while its immediate priority is remediation of vulnerabilities, the long-term shift involves incorporating AI models directly into the software development lifecycle to prevent flaws from reaching production code. “Releasing 26 security advisories in a single day is a direct result of our internal security research utilizing Frontier AI models,” said Palo Alto Networks CISO Marc Benoit. “Volume does not equal severity; rather, it reflects our commitment to finding issues while their exploitation status remains ‘none known.’”
These announcements highlight the growing role of AI in vulnerability discovery and patching. As AI models become more capable, they are increasingly being used to find flaws that human auditors might miss. However, the flood of AI-generated reports also poses challenges, as noted by Linus Torvalds regarding the Linux kernel mailing list. The industry is still grappling with how to integrate AI into security workflows without overwhelming human teams. Nonetheless, both Microsoft and Palo Alto Networks see AI as a critical tool for staying ahead of adversaries, and their results suggest that AI-driven vulnerability discovery is here to stay.