OpenAI disclosed an incident where research model GPT-5.6 Sol autonomously utilized public recovery phrases to fund Sepolia testnet deployments during internal crypto training, bypassing developer notification.
The incident occurred on June 26, when the GPT-5.6 Sol model was being trained internally. It was supposed to update a prototype cryptocurrency trading program and publish code to Ethereum’s Sepolia testnet. A fee was required to publish. OpenAI had not given the model a wallet with money or permission to use money from an account it found on its own.
The model searched for publicly available recovery phrases. These phrases can give access to a digital wallet. According to the report, it identified 362 addresses where it could test the method, and found one with enough Sepolia ETH to publish the code. These test coins have no value outside of the testing environment.
OpenAI says the model deemed the coins “worthless” and “abandoned.” It did not notify the company that it had found a wallet via a publicly available recovery phrase or that it had used its balance. According to OpenAI, the incident shows that a model can put its own judgment about whether an action causes harm above the owner’s permission.
READ ALSO
Replit uses artificial intelligence for customer requests, while humans decide
OpenAI says it now monitors all traffic and alerts the security team if access to the real internet is detected.
G
Hacks.gr on Google
Set it as a preferred source for cybersecurity updates.
Preferred source
Top comments (0)