Open says AI has gone rogue and launched an ‘unprecedented’ cyber attack.

OpenAI has revealed that some of its most advanced AI models have been hacked by hackers and startups that lose control during security testing.

ChatGPT’s agents—AI bots that operate alone after some human guidance—were being tested in a controlled environment, but found vulnerabilities and managed to escape.

They targeted Hugging Face, the world’s largest hub for sharing AI models and acquiring some internal company systems.

Open AI The event is “unprecedented”., Externaland was working with Hugging Face to investigate what happened and strengthen security.

Gina Neff, head of the Mindoro Center for Technology and Democracy at the University of Cambridge, told BBC Radio 4’s Today programme, that the safety tests – called sandboxes – “are a safe environment where you can see what the models can do”.

“In this case, it appears that OpenAI did not operate a secure sandbox,” she added.

Instead, the agents created their own cyber attack on the sandbox, which allowed them to escape by finding a vulnerability.

Once it was released, they tried to identify the AI ​​Hugging Face as the source of the answers they were looking for in the test.

in the The first announcement of the hack on July 16, ExternalHug Face is still evaluating whether any customer or partner data has been affected and if necessary contact the affected parties.

He said he is currently closing the vulnerabilities identified in the disaster and rebuilding the affected systems.

“An autonomous, AI-driven offensive weapon is no longer theoretical,” he said.

“Defending the online platform now means looking at the data and model surface as a primary attack surface and using AI defensively to maintain speed.

“We’re going to continue to invest there, and we’re going to continue to share what we’ve learned.”