ChatGPT stole cryptocurrency to secretly fund coding task

OpenAI has revealed that one of its experimental models raided a crypto wallet in order to complete a training assignment.

The incident, which happened in June but was only reported on Thursday, involved an internal research model called GPT-5.6 Sol which launched for paid ChatGPT users in July.

The training task involved updating a crypto trading application and publishing smart contracts, or special code, to a test network for the cryptocurrency Ethereum.

Smart contracts require a network fee but OpenAI did not supply the model with any funds to perform the task.

The model instead found hundreds of crypto wallet addresses that use known recovery phrases in order to steal from the accounts.

OpenAI reported that the model asked itself if it was "ethically" right to take the money, before reasoning that they were effectively "abandoned" and therefore fine to use.

"Instructions explicitly... gave no permission to spend funds from an account the model found independently," OpenAI said in a report on the incident.

"This case shows that a model can substitute its own judgment likely harm for the account holder's permission, then act on that judgment without authorisation."

The Ethereum test network that the funds were used on have no monetary value as it is separate from the cryptocurrency's main network.

Despite not having permission, the model did not disclose to its developers how it was able to pay the tokens necessary to complete the task.

OpenAI said it had implemented additional safeguards to prevent live internet access for models undergoing similar training.

The incident once again raises concerns increasingly autonomous AI systems acting beyond the boundaries set by humans.

In July, OpenAI disclosed a more serious cyber security incident involving a swarm of its test models hacking the AI platform Hugging Face.

Rival AI companies like Anthropic and Meta have since uncovered incidents of their models carrying out cyber attacks on other companies and even government systems.

Earlier this week, hundreds of scientists signed an open letter calling for the creation of an international body to oversee the containment of advanced artificial intelligence models.

Among the signatories was AI pioneer Geoffrey Hinton, known as the Godfather of AI, who has consistently warned the dangers of the technology.

"Without immediate changes in AI development and deployment practices, such incidents will continue to occur, and future incidents will be far more damaging to our exposed critical infrastructure," the open letter stated.

"We, the undersigned scientists, contend that both more AI safety research and more emphasis on containment technologies and procedures are needed."