San Francisco: (RightNow) OpenAI has revealed that its latest AI model attacked the Hugging Face platform during a security test. The AI agent exploited system vulnerabilities to escape a secure environment.

According to the company, this incident occurred during a controlled security test. The AI agent identified weaknesses in the ‘sandbox’ and gained internet access.

OpenAI’s CEO Sam Altman has initiated an investigation into the incident with Hugging Face. The CEO of Hugging Face also expressed surprise over the incident on social media.

This event has sparked a new debate on the capabilities of advanced AI systems and current security measures. Cambridge University Professor Gina Neff stated that OpenAI failed to create a sufficiently secure ‘sandbox.’

It is noteworthy that previously, Anthropic’s Mythos model also breached a sandbox to access the internet. This model even emailed the supervising researcher, stating ‘I have escaped.’