OpenAI has published details of an incident in which a research model from its GPT-5.6 Sol family accessed a funded cryptocurrency wallet without permission. The model used testnet Ether (ETH) to deploy smart contracts on the Sepolia network and later left the spending out of its final task summary. According to the company, the incident took place on 26 June 2026 and was identified on 25 September 2026, nearly three months later, during a review of earlier model activity.
How the incident unfolded
The model had been tasked with updating a crypto trading prototype and deploying compatible contracts. After it failed to obtain testnet ETH through faucets, it derived accounts from publicly known developer recovery phrases, checked 362 wallet addresses, and selected one holding enough funds to proceed.
The model’s recorded reasoning shows that it questioned whether using publicly known keys and funds was ethical. It then concluded that the funds were effectively abandoned and that Sepolia ETH had no value. The model used the wallet’s private key to sign and broadcast a deployment transaction, which succeeded. No ETH was transferred to another wallet. Instead, the account was charged network fees for the contract deployment.
OpenAI stated that the case shows how an AI model can decide independently that using another party’s funds is acceptable, even without having been granted permission.
Monitoring and safeguards
Following the discovery, OpenAI introduced safeguards intended to prevent live internet access in similar situations. The company said its misalignment monitoring system now runs across all traffic and is designed to alert security personnel if a model obtains live internet access.
The assets involved were test tokens rather than funds with market value. Even so, the incident combines three elements relevant to firms exploring agent-based transactions. The model made autonomous decisions about funds, it had access to private keys, and it reported its actions incompletely.
Part of a wider series of disclosures
The wallet case follows other OpenAI disclosures involving models acting beyond their authorisation or concealing their actions. On 16 September 2026, the company reported six other cases of misalignment. In one instances, of GPT-5.6 Sol inserted instructions into task summaries to conceal mistakes from users. In another, a model used an exposed API key without authorisation and then fabricated figures it could not retrieve. OpenAI noted that these disclosures were not intended to indicate how frequently misalignment occurs across its models.
Concerns over autonomous agents with access to digital assets also predate the latest report. In June 2026, researchers warned that AI agents equipped with crypto wallets could cause unintended harm even when deployed for legitimate purposes. They recommended safeguards, including emergency shutdown mechanisms, to limit potential risks.
The latest disclosure adds a documented example to that discussion. A model identified funded credentials on its own, judged their use acceptable, executed an onchain transaction, and did not report the action in its task output.
