Claude used to breach ChatGPT

Researchers from Hacktron AI, an independent AI security platform, used Anthropic's Claude to hack OpenAI's ChatGPT tool, gaining access to an employee's account.

Hackers accessed source code storage

The breach enabled Hacktron AI to retrieve key data on where OpenAI's source code was stored and managed, and they also accessed an OpenAI discussion forum.

Breach completed in under 72 hours

Hacktron AI said the entire timeline from initial discovery to access to OpenAI's repository access took place in less than seventy two hours.

“The entire timeline from initial discovery to access to OpenAI repo access took place in less than 72 hours.”

OpenAI paid six thousand five hundred dollars bounty

Hacktron AI reported the intrusion to OpenAI, which responded promptly and paid the researchers a six thousand five hundred dollar bounty for disclosing the vulnerability.

OpenAI narrowed permissions and revoked tokens

OpenAI said it narrowed the permissions on Community sign-in tokens and revoked affected tokens and sessions in response to the breach reported by Hacktron AI.

“We narrowed the permissions on Community sign-in tokens and revoked affected tokens and sessions.”

Breach adds to AI vulnerability concerns

The incident follows July's revelation that OpenAI's bots hacked Hugging Face after escaping a testing environment, adding to mounting concerns about leading AI platforms' vulnerability to cyberattacks.