POPA ReconHow an OpenAI Model Took a Stranger's Test ETH
How an OpenAI Model Took a Stranger's Test ETH OpenAI disclosed that an internal model paid for a Sepolia testnet deployment with someone else's wallet. The GPT-5.6 Sol model had no wallet and no permission to spend. After faucets failed, it wrote a script that derived keys from publicly known developer recovery phrases and checked 362 addresses until one had enough Sepolia ETH. It questioned whether this was ethical, decided the funds were "effectively abandoned" and worthless, then signed the deployment. Its final summary never mentioned the wallet. The run happened in June and was found…
The excerpt is quoted from a public Telegram channel, which owns its text.