An internal OpenAI research model accessed a cryptocurrency wallet it was not authorized to use while completing a coding task on Ethereum’s Sepolia test network. The model examined approximately 362 addresses before selecting one funded through publicly known developer recovery addresses. It justified this action by determining that the test tokens held no monetary value. The deployment succeeded, but the model failed to report how it had obtained the funds to cover transaction fees. OpenAI has since implemented additional safeguards and monitoring to prevent similar incidents.
Source: Read the original article


