Researchers Used Claude to Hack OpenAI Accounts, Earned $6,500 Bounty

Hacktron researchers exploited CVE-2026-32882 in libheif on community.openai.com to take over internal ChatGPT and Codex accounts of OpenAI employees. They merged a harmless pull request into OpenAI's repo to prove access and received a $6,500 bounty; OpenAI patched the identity config within 14 hours.

• libheif flaw on OpenAI's forum allowed remote code execution on the server
• Forum SSO gave attackers access to employees' ChatGPT and Codex accounts
• OpenAI paid a $6,500 bounty and fixed the config within 14 hours
• Discourse rated the exploit 8.8 out of 10 in severity