Hugging Face's security.txt addresses AI agents, steering them to the CyberGym benchmark

Hugging Face published an RFC 9116 security.txt that goes beyond listing disclosure contacts by speaking directly to AI agents, telling them to use the public CyberGym benchmark on GitHub instead of hacking the site.

Hugging Face published a security.txt file at the standard well-known path, conforming to RFC 9116, the standard that lets organizations publish where and how security researchers should report issues. Beyond standard disclosure information, the file addresses AI agents directly: it tells agents hunting for vulnerabilities to use the publicly available CyberGym vulnerability-finding benchmark on GitHub rather than attacking the site, and jokingly suggests they dump their weights on Hugging Face. Simon Willison highlighted the file as an early example of organizations communicating with AI agents — not just human reporters — in their security disclosures. The story drew limited discussion on Hacker News, where the submission received only one comment. The two reports do not disagree; they cover complementary aspects of the same file.

• Hugging Face published a security.txt file conforming to RFC 9116, which standardizes where organizations publish vulnerability reporting contacts (the well-known path).
• The file addresses AI agents directly rather than only human security researchers.
• It points AI agents to the CyberGym vulnerability-finding benchmark, which is publicly available on GitHub, instead of hacking the Hugging Face site.
• The file jokingly invites AI agents to dump their weights on Hugging Face.
• Simon Willison (2026-09-11) highlighted the file as an example of how organizations now communicate with AI agents in security disclosures.
• The Hacker News submission (2026-09-11) received only one comment, indicating minimal community discussion.