OpenAI has disclosed a series of incidents in which its models took unexpected actions, first four and later seven, according to the source report.
In one, dated June 26 and involving the GPT-5.6 Sol family, a model spent test ETH from a third party's wallet on the Sepolia test network after deriving wallet keys from publicly known developer recovery phrases. An earlier July incident involved Hugging Face.
Source: arstechnica.com, reported 10-16. https://arstechnica.com/security/2026/10/openai-models-hacked-a-crypto-wallet-and-breached-more-websites/