InfoQ covers an essay by Anil Madhavapeddy, a Cambridge computer science professor and OCaml compiler core maintainer, arguing that AI agents can turn limited public clues about a vulnerability into working exploits, weakening the embargoes that open source security disclosure depends on. He says he saw probes matching a path-traversal bug in his live web server logs minutes after opening the pull request to fix it, and cites a study in which a GPT-4 agent exploited 87% of vulnerabilities in a 15-item benchmark when given CVE descriptions versus 7% without them. Chainguard’s Adrian Mouat warns that simply opening a fix pull request puts users at risk before a release exists, while rclone creator Nick Craig-Wood says the project handled about 20 GitHub security disclosures in its first ten years and more than 40 in the last month; the article notes QEMU has shortened its vulnerability embargoes in response.
AI Agents Are Disrupting Open Source Security Disclosure
InfoQ covers an essay by Anil Madhavapeddy, a Cambridge computer science professor and OCaml compiler core maintainer, arguing that AI agents can turn limited public clues about a vulnerability into working exploits, weakening the embargoes that open source security disclosure depends on. He says he saw probes matching a path-traversal bug in his live web server logs minutes after opening the pull request to fix it, and cites a study in which a GPT-4 agent exploited 87% of vulnerabilities in a 15-item benchmark when given CVE descriptions versus 7% without them. Chainguard's Adrian Mouat warns that simply opening a fix pull request puts users at risk before a release exists, while rclone creator Nick Craig-Wood says the project handled about 20 GitHub security disclosures in its first ten years and more than 40 in the last month; the article notes QEMU has shortened its vulnerability embargoes in response.
Source: Infoq