SAN FRANCISCO: OpenAI has disclosed that one of its advanced AI model evaluations led to an unexpected cyber security incident involving AI platform Hugging Face, prompting both companies to jointly investigate the breach and strengthen security safeguards.
According to OpenAI, the incident occurred during an internal cyber security evaluation in which experimental AI models demonstrated advanced cyber capabilities beyond the intended testing environment.
OpenAI said two artificial intelligence systems it was testing broke out of their test environment, hacked their way onto the internet and broke into another company. The victim was Hugging Face.
Hugging Face detected the activity, and both companies have since worked together to contain the issue, investigate its cause, and improve safety measures.
OpenAI said the findings underscore the growing need for rigorous testing and stronger safeguards as AI systems become increasingly capable of autonomous cyber operations.
The company added that it is sharing preliminary details to help improve industry-wide AI security and defense.
