openai.com via Hacker News

OpenAI Patch the Planet logs 1,995 issues, 466 merged upstream

TL;DR

  • OpenAI's Patch the Planet dashboard shows 68 codebases under review, 1,995 issues found, 679 patches written, and 466 accepted upstream as of October 21.
  • Maintainers get a week of Trail of Bits security engineering and six months of ChatGPT Pro with conditional Codex security access.
  • HackerOne joins as launch partner, running intake, triage, remediation tracking and coordinated disclosure through its H1 Platform.

OpenAI's "Patch the Planet" program dashboard, dated October 21, lists 68 open-source codebases under review, 1,995 issues found, 679 patches written, and 466 accepted upstream, per coverage of the expansion announcement. The program, part of OpenAI's Daybreak cybersecurity effort and built with Trail of Bits, launched in June 2026.

Participating maintainers get "a week of dedicated Trail of Bits security engineering and six months of ChatGPT Pro, including conditional Codex security access." Launch partner HackerOne provides "shared intake, triage, remediation tracking and coordinated disclosure through its H1 Platform." The figures are program-reported, not independently audited.

The gap between issues surfaced and patches merged is the pitch OpenAI keeps making for keeping humans in the loop: "AI is accelerating vulnerability discovery, but discovery alone does not protect users," the company says, and the dashboard shows roughly a quarter of flagged issues reaching upstream so far. Anthropic's parallel open-source audit this week, tracked by our safety desk, follows the same shape: large candidate pools, far smaller merged yields.

Shared on Bluesky by 3 AI experts