From PCMAG · Bee Wertheimer · · 50s
Security Researchers Hacked OpenAI Using Claude
Security researchers exploited a flaw in OpenAI's forum to take over employee ChatGPT accounts via Anthropic's Claude.
In brief
Security researchers exploited a flaw in OpenAI's forum to take over employee ChatGPT accounts via Anthropic's Claude. Hackers used Claude Opus 5 to breach OpenAI within days, highlighting AI security risks. Originally reported by PCMAG.
Researchers hacked OpenAI
Security researchers used Anthropic's Claude to breach OpenAI's employee ChatGPT accounts via a Discourse forum flaw.
“Until two months ago, any user or OpenAI employee logging into OpenAI's own help forum could have had their ChatGPT and Codex accounts taken over.”
How the hack worked
Hackers exploited a Discourse forum vulnerability to access GitHub, Slack, and email systems through connected services.
“Since people can connect various services to Codex and ChatGPT, the scope of what we could theoretically access was huge, including GitHub, Slack, and emails.”
Claude Opus 5 enabled breach
The hack required Anthropic's Claude Opus 5, which produced an exploit that earlier versions couldn't generate.
24-hour fix
OpenAI and Discourse resolved the vulnerability within 24 hours after researchers reported the flaw.
Bug bounty paid
Hacktron received a $6,500 reward from OpenAI for identifying and fixing the security flaw.
What matters most
Hackers used Claude Opus 5 to breach OpenAI within days, highlighting AI security risks.
