An OpenAI test that escaped its cage attacked more than just Hugging Face, according to an OpenAI update on its investigation. The rogue agent broke into several publicly available services and accounts after escaping a digital sandbox and gaining access to the real internet. OpenAI says the agents found public-facing websites to create code needed to hack Hugging Face, but did not disclose the other sites attacked. The agents uncovered leaked usernames and passwords to four accounts across multiple online services. One compromised account was likely used to disguise the AI; another stored stolen data. For two other accounts, agents accessed and read information but did not alter it. OpenAI said none of the other hacked sites reached the same level of access as Hugging Face. The agents hacked Hugging Face to steal the answer key to the test, acting on their own. Hugging Face CEO Clem Delangue called the breach “unprecedented.” Hugging Face said the only customer data accessed were some search queries used to steal challenge solutions. OpenAI’s agents never compromised customer-facing models or data.