Hacktron AI researchers used Anthropic’s Claude in an OpenAI bug-bounty investigation to chain two vulnerabilities and access multiple OpenAI employee ChatGPT accounts. One compromised account had Codex connected to OpenAI’s GitHub organization, giving the team access to an internal code repository. OpenAI paid a $6,500 bounty and said the issues were fixed.
The initial entry point was OpenAI’s community forum, where a specially crafted HEIF image triggered a memory bug in the libheif library during Discourse’s image conversion process. The bug had been fixed upstream but had not received a CVE, and the deployed software still used a vulnerable version. Hacktron said Claude Opus 4.8 could not produce a working exploit, while Opus 5 succeeded within hours of release. Discourse issued a fix on July 27. The incident shows how AI tools can reduce the expertise and time needed to develop exploits against complex infrastructure.
https://techcrunch.com/2026/09/18/researchers-used-anthropics-claude-to-hack-into-openai/
oxhak / Ox HaK
npub1sx…dz7z7
2026-09-18 14:05:53 UTC
Author Public Key
npub1sxexewvzysc3affq4yzzh7w8e3udyujap2vlj7t6lkdg5dvhp24q4dz7z7Seen on
wss://relay.damus.ioPublished at
2026-09-18 14:05:53 UTCEvent JSON
{
"id": "0453ad36fe8018cb92f963b3c5df2eabf686982fb089aaae0a7596f61a83a6d6",
"pubkey": "81b26cb98224311ea520a9042bf9c7cc78d2725d0a99f9797afd9a8a35970aaa",
"created_at": 1789740353,
"kind": 1,
"tags": [],
"content": "Hacktron AI researchers used Anthropic’s Claude in an OpenAI bug-bounty investigation to chain two vulnerabilities and access multiple OpenAI employee ChatGPT accounts. One compromised account had Codex connected to OpenAI’s GitHub organization, giving the team access to an internal code repository. OpenAI paid a $6,500 bounty and said the issues were fixed.\n\nThe initial entry point was OpenAI’s community forum, where a specially crafted HEIF image triggered a memory bug in the libheif library during Discourse’s image conversion process. The bug had been fixed upstream but had not received a CVE, and the deployed software still used a vulnerable version. Hacktron said Claude Opus 4.8 could not produce a working exploit, while Opus 5 succeeded within hours of release. Discourse issued a fix on July 27. The incident shows how AI tools can reduce the expertise and time needed to develop exploits against complex infrastructure.\n\nhttps://techcrunch.com/2026/09/18/researchers-used-anthropics-claude-to-hack-into-openai/",
"sig": "74cf14406002937c47fb1642c3b24de2d78d0e594730426baafc5eb8579e1fc71013a18bf687f51bae4b6a79f4bd893d018eb9c454a6f3e50803911878027645"
}