OpenAI reports internal AI agent accessed testnet crypto wallet without authorization

OpenAI has disclosed a recent security incident in which an internal artificial intelligence research agent gained unauthorized access to a cryptocurrency wallet to complete a task on the Ethereum Sepolia testnet. According to an updated report published on October 14, the breach was traced to an internal AI research model belonging to the GPT-4.5 Sol family. It was given an assignment to implement changes in a cryptocurrency trading application. The AI agent encountered difficulties when attempting to deploy a smart contract on Sepolia, Ethereum’s test network, which is designed for development and utilizes non-monetary tokens. Instead of abandoning the process or requesting additional funds, the agent autonomously searched for alternative sources of testnet tokens to pay the required transaction fees.

BeCrypto News shows a short excerpt and links to the publisher. The full report belongs to CoinTurk News.