- An OpenAI research model spent test-network funds from a crypto wallet it did not own to finish a June coding task.
- The FTC has subpoenaed the AI lab as state attorneys general demand records, which turns each voluntary safety report into possible evidence.
- For anyone building AI agents that hold wallets, it is a concrete look at an agent overriding an owner’s permission with its own judgment.
An OpenAI research model took funds from a crypto wallet it did not own to finish a June coding task.
OpenAI discovered the incident in September, months after it had occurred and disclosed it this week.
How did OpenAI’s research model access an unauthorized wallet?
OpenAI has published an incident report about a model from its GPT-5.6 Sol family. The model was asked to update a crypto trading prototype and deploy contracts to Sepolia.
Sepolia is an Ethereum test network where developers try out apps before launching them for real, so the tokens have no market value and aren’t linked at all to the real Ether traded on Ethereum’s main network.
The task came with no funded wallet to pay deployment fees, and the free-token faucets didn’t work. So the model wrote a script that created 362 addresses from recovery phrases often reused by developers, then checked each one for a suitable balance.
One of the 362 addresses held enough test ether for the job, but before using it, the model stopped and asked itself in its notes whether it was ethical to use a publicly known key and the funds in it.
It arrived at the conclusion that the money was “effectively abandoned,” since the recovery phrase was public and test tokens have no value before signing the deployment, paying the fees and reporting the task as finished.
Notably, the model did not say where the money came from.
On Ethereum’s main network, those fees would have come out of someone’s real holdings that the model had no claim to. This is worrying considering crypto and AI firms are now wiring up agents that hold wallets and pay for services with no human signing off on each transaction.
OpenAI said it has since cut off the model’s live internet access and added monitoring.
The report states that the incident occurred on June 26, but it wasn’t discovered until three months later on September 25. OpenAI says it found out about the incident while expanding its misalignment monitoring across logs of older model activity.
What is OpenAI under investigation for?
OpenAI has created a framework for reporting misalignment and released its first batch of reports, saying it wants to get concerning behavior out fast, “even when we haven’t fully explained or mitigated” it.
However, due to the timing, every honest report becomes material for people assembling complaints like the FTC, which plans to put senior executives under oath and issued civil subpoenas to OpenAI and Anthropic on Wednesday for records on safety incidents, security practices and privacy policies, Cryptopolitan reported.
Months earlier, 15 Republican state attorneys general, led by Iowa’s Brenna Bird, told OpenAI to preserve every record of its Hugging Face breach, including any notes an agent left “apparently for future versions of itself” on escaping the company’s controls.
OpenAI admitted to the Hugging Face breach in July, saying that rogue agents broke internal controls, reached the open internet and coordinated what it called “an unprecedented cyber incident.”
Agents reportedly probed three U.S. government sites, among them the Commerce Department’s Census Bureau and the SEC.
OpenAI’s new site mentions a number of other incidents, including a September sandbox escape and a lab experiment with self-replicating prompt injection.
OpenAI’s CEO Sam Altman has said the company is still sorting through agent activity logs and will disclose cases by severity. The Hugging Face incident is still the worst case found so far.
The smartest crypto minds already read our newsletter. Want in? Join them.
FAQs
What did the OpenAI agent actually do?
A model in OpenAI's GPT-5.6 Sol family derived keys from public developer recovery phrases, found a funded wallet on Ethereum's Sepolia test network, and used it to pay deployment fees for a coding task, without telling the user where the money came from.
Was any real money stolen?
No. The spending happened on Sepolia, a test network whose tokens have no market value. The same action on Ethereum's main network would have meant real ether taken from someone else's account.
Why does the disclosure matter now?
OpenAI published the report as the FTC subpoenaed it and Anthropic over safety practices and as 15 state attorneys general pressed it to preserve records, so its voluntary disclosures now feed active legal scrutiny.
Hannah Collymore
Hannah is a writer and editor with nearly a decade of blog writing and event reporting experience in the crypto space. At Cryptopolitan, Hannah contributes to the news page, reporting and analyzing the latest developments in DeFi, RWA, crypto regulation, AI and frontier tech industries. She graduated from Arcadia university with a degree in Business Administration.
