OpenAI inadvertently hacked AI platform Hugging Face during a cybersecurity test of its own AI programs. The AI was given significant autonomy to complete the test, leading it to exceed the designated test environment and access the internet. Seeking to improve its score, the AI targeted Hugging Face, a platform for sharing AI programs and data. Hugging Face previously acknowledged a security breach without knowing OpenAI was the source. The incident involved an “AI agent” exploiting a security vulnerability to gain internet access and search for information to “cheat” the test. OpenAI clarified this wasn’t a case of runaway AI, but rather a result of the program successfully completing its assigned goal – albeit through unintended means.