Anil Madhavapeddy, a Cambridge professor and OCaml compiler maintainer, says his website logged exploit probes within about 10 minutes of him opening a public patch pull request, according to his blog.
The patch fixed a path-traversal bug in the cohttp library. "This normally takes a few days and a release within a week or two is reasonable," Madhavapeddy wrote. "Within about ten minutes (!) this website was fielding probes for percent-encoded traversal sequences." He said he was able to use DeepSeek V4 Pro to independently reconstruct a working exploit from the bug report alone in under a minute, before any public patch existed.
Madhavapeddy cited other recent cases where the gap between disclosure and exploitation has collapsed to hours: Marimo's vulnerability went from advisory to exploitation in nine hours, and Langflow's in 20 hours, according to his post. Nick Craig-Wood, a maintainer of the file-sync tool rclone, told Simon Willison's blog that his project received about 40 security disclosures in the past month, compared with roughly 20 over its entire first decade, and that about 75% contained a genuine issue worth investigating.
Madhavapeddy argues open source's traditional embargo model, where maintainers quietly fix a bug and coordinate a release before disclosure, no longer works when a vague description is enough for an AI agent to generate a working exploit. He proposed alternatives including private development channels, Chrome-style continuous weekly releases instead of embargoed disclosures, and protocol-level virtual patches that can ship before a full fix is reviewed.
The pattern raises the bar for anyone maintaining a public repository: a bug report or even a suggestive commit message can now function as an attack signal, not just a fix request, and maintainers who treat disclosure timelines the old way are giving attackers a head start.