OpenAI AI agent goes rogue? Crypto funds used without permission
- Author:
- Publish date:
OpenAI has disclosed a incident in which one of its internal AI agents used a crypto wallet without the owner's permission.
According to an X post, the agent was working on a task on Ethereum's Sepolia testnet when it ran short of gas, the fee needed to complete a transaction. Sepolia tokens carry no monetary value, so no real funds were affected.
In the OpenAI alignment blog, the company stated, “An internal research model asked to update a cryptocurrency trading application used funds from a wallet that the task had not supplied to pay for a deployment on Sepolia, an Ethereum test network. It derived wallet keys from publicly known developer recovery phrases, found an account with test-network funds, and used that account to pay for the requested contract deployment.”
Agent weighed the ethics, then proceeded
The agent was asked to update a crypto trading app and deploy smart contracts, but it was not given a funded wallet.
After its attempts to get free test tokens failed, it searched for funded wallets using publicly known recovery phrases. It checked 362 addresses and found one with enough tokens.
Recommended Articles
Logs showed the agent questioning whether using the funds was ethical. It then called them "effectively abandoned," because anyone could access the wallet through its public recovery phrase and Sepolia test tokens have no real value. It then signed the transaction.
The contracts were deployed. The agent later reported the task as complete without saying it had used someone else's wallet.
OpenAI tightens controls
The incident occurred on June 26 and was identified on Sept. 25, and later updated on Oct 15. OpenAI has since added stricter monitoring and tighter internet-access restrictions for its research models.
“We discovered this incident while extending our misalignment monitoring to records of earlier model activity. We have implemented additional safeguards to prevent live internet access for cases like this. We now run our misalignment monitoring system on all traffic which is designed to page security if live internet access were to be obtained,” OpenAI wrote in the blog post.
Though no real ETH was involved, the case points to a security risk as autonomous agents gain access to blockchain infrastructure.
Using someone else's funds without approval remains a serious issue, even on a test network.
