Maintainers report exploit attempts within minutes of bug disclosure

Anil Madhavapeddy, a Cambridge computer science professor and core OCaml maintainer, reports that when a security patch was shared for discussion, his project's website was fielding probes for percent-encoded traversal sequences within about ten minutes, well under the days-to-weeks window that patch review and release normally take. He attributes this to automated watchers on public repositories and coding agents capable of turning a vague hint of a bug into a working exploit attempt; he says he demonstrated similar capability himself using agents, switching tools when one declined the task. Separately, rclone maintainer Nick Craig-Wood confirms a related but distinct effect in Hacker News comments: after receiving roughly 20 security disclosures in the project's first ten years, rclone received over 40 in a single recent month, about 75% containing something worth investigating, and GitHub's CVE assignment turnaround has slowed from 2-3 days to 3-4 weeks under this load. Both accounts are practitioner reports rather than controlled studies, with no independent reproduction, dataset, or methodology disclosed beyond the named maintainers' own observations.

rss · Simon Willison · Aug 28, 22:12

「The assumption being tested」 Open source security workflows have long relied on coordinated disclosure and embargo periods, giving maintainers days to weeks to develop and release a fix before details become public. That timeline assumed attackers needed comparable time to notice a disclosed issue and build a working exploit, a gap that automated code analysis agents appear to be closing.

「Who this affects」 This is most relevant to maintainers of public open source projects who discuss patches or security issues in the open before release, particularly on widely watched repositories. Organisations that consume such projects should check whether their upstream dependencies follow embargoed disclosure practices that assume a multi-day response window, since that assumption is what the reports call into question. The rclone maintainer's account also points to a secondary, more concrete exposure: CVE assignment delays of 3-4 weeks mean downstream consumers may see fixed releases shipped with CVE-PENDING status, complicating vulnerability tracking and patch prioritisation in the interim.

「What reduces the risk」 No fix exists for the underlying dynamic; both maintainers frame this as a need to redesign disclosure and embargo processes rather than a bug to patch. In the meantime, projects report using AI tools themselves to triage the resulting surge of disclosures and speed up fix development, and downstream consumers can compensate by not relying on CVE assignment speed as a signal and instead tracking upstream advisories and patched releases directly.

Tags: #security, #vulnerability-disclosure, #AI-agents, #open-source-maintenance