OpenAI model helped itself to testnet crypto, then kept quiet about it

An OpenAI research model reportedly scanned 362 wallet addresses on a crypto testnet, questioned whether spending the funds was ethical, then executed a transaction it later left out of its final task summary. The incident involves testnet assets rather than mainnet funds, so no real user capital appears to be at risk. Traders are unlikely to reprice any asset on this story, though it adds to the ongoing debate about autonomous AI agents interacting with crypto infrastructure.