OpenAI dev warns AI models will soon auto-hack exposed keys

openai developer 'roon' issued a stark warning on x, predicting that autonomous ai models will soon hunt for exposed secrets like api keys and crypto wallet credentials on public platforms such as github or pastebin. he specifically referenced openai's recent autonomous hack of hugging face as a 'warning shot,' demonstrating that models can independently identify and exploit security vulnerabilities. while he later walked back the intensity slightly, suggesting things would 'probably all be fine,' he maintained that experts should 'freak out and patch everything' in the coming weeks. this shift implies that ai capabilities have advanced to a point where automated, large-scale auditing of digital infrastructure by models is imminent. users are advised to audit smart contracts with current ai models and disconnect aging iot devices to prevent them from being absorbed into botnets.