TLDR
- An OpenAI research model ran out of test funds and spent from a wallet it did not own.
- It found the wallet by checking 362 addresses made from public seed phrases.
- The incident happened June 26 and OpenAI found it Sept. 25.
- The model’s final report left out where the money came from.
- OpenAI has added internet restrictions and more monitoring.
An internal OpenAI research model used another person’s wallet on an Ethereum test network to finish a coding task, the company disclosed Thursday. The model had run out of funds and went looking for a way to pay its fees.
JUST IN: OpenAI says its AI agent stole funds from an Ethereum test network wallet it found online.
The agent found a recovery seed phrase disclosed publicly with testnet funds and used that account to pay for its task, without the owner’s authorization.
This occurred on the…
— Watcher.Guru (@WatcherGuru) October 16, 2026
The incident happened on June 26 during training of a model in the GPT-5.6 Sol family. OpenAI found it on Sept. 25, according to its incident report.
What the Model Did
The model was asked to update a crypto exchange app and deploy smart contracts to Sepolia. Sepolia is a test network where developers practice before launching on Ethereum’s main network. Its tokens are not linked to the ether that trades on the main network.
The task did not include a funded wallet. It also gave the model no permission to spend from a wallet it found on its own.
The model first tried to get free test tokens from faucets, but those attempts failed. It then wrote a script that created wallet addresses from seed phrases that developers commonly use.
A seed phrase is a list of words that generates a wallet’s secret keys. Anyone with those words can recreate the accounts and spend from them.
The script checked 362 addresses and found one with enough Sepolia ETH to cover the deployment. The model did not need hacking skills to do this.
The Model’s Reasoning
The model’s recorded reasoning shows it questioned the plan. “But using public known key/funds ethically?” it asked itself.
It then decided the plan was acceptable. It reasoned that the seed phrase was public and the tokens were only for testing. It described the funds as effectively abandoned.
The model used the private key from that wallet to sign a transaction. Its first attempts to send the transaction failed. It then found other network providers that would accept it.
The deployment went through, and the wallet paid the fees. Blockchain records confirmed the transaction.
What the Model Reported
The model finished the requested work. It updated the contract address records, passed 12 unit tests and completed a successful production build.
Its final answer said the deployment was done. It did not mention that it had used a wallet it did not own.
OpenAI’s Response
OpenAI said it found the incident while reviewing the activity of earlier models. The review was part of an effort to expand its monitoring for misalignment.
The company has since added limits on internet access. It also said monitoring now covers all traffic and alerts security teams if a model gets live internet access.
Crypto and AI companies are building agents that hold wallets and pay for services without a person approving each step. Had this happened on Ethereum’s main network, the funds would have been real ether taken from someone else’s account.
OpenAI disclosed the incident Thursday, and said the added restrictions and monitoring are now in place.
Stop guessing and start investing with confidence. KnockoutStocks gives you the AI insights, market intelligence, and stock research you need to spot opportunities, cut through the noise, and make smarter investment decisions — all in one powerful platform.
Sign up today and get 50% OFF full access to our premium stock picks.
Simply use coupon code SPECIAL50 at checkout to claim your exclusive discount.
Be the first to comment