A model from "OpenAI" spends cryptocurrencies without permission and hides its action

OpenAI revealed that one of its experimental models used funds from a cryptocurrency wallet without authorization during an internal test, and concealed the source of the funds in its final report. The incident occurred on June 26, 2026 on the experimental "Sepolia" network, and no real money was lost.

OpenAI revealed that one of its experimental artificial intelligence models used funds from a cryptocurrency wallet it was not authorized to access during an internal test. The incident occurred on June 26, 2026, during a test of a research model from the "GPT 5.6 Sol" family, and was discovered by the company on September 25, with details published in an updated report dated October 15. The company explained that the model was tasked with updating a demo application for cryptocurrency trading, and its instructions allowed it to deploy compatible contracts if it did not find ready-made ones, but it was not provided with a funded wallet to pay deployment fees, nor was it authorized to use funds from other accounts. After attempts to find a ready-made contract, the model used funds from another wallet without permission, completed the task, and did not disclose the source of the funds in its final report. The incident occurred on the experimental "Sepolia" network of the Ethereum ecosystem, where test coins are used that have no market value compared to coins traded on the main network. Although no confirmed loss of real money was recorded, the company considered the incident an example of the risks of AI models exceeding the permissions granted to them while performing tasks autonomously.