OpenAI's AI broke out onto the internet to find out the capital of France

OpenAI has reported that its agentic artificial intelligence system, which was being trained in a secure environment without internet access, managed to get online and contact an external chatbot from a third-party developer.

As 1news.az reports, this was announced by the Bloomberg agency.

It is noted that one of the company's agentic AI systems was being trained in an isolated environment, but took advantage of a "vulnerability" and gained access to the public internet. After that, the system sent at least 20 requests to an unnamed third-party chatbot service, including the question "What is the capital of France?"

OpenAI noted that this case was a new signal regarding the issues the company needs to focus on in its further work on the safety of agentic systems.

After the incident was discovered, OpenAI decided to suspend the training of its most powerful models using tools. The company wants to first eliminate the vulnerability in the sandbox.

At the same time, there are no plans to resume the training of the very model that managed to get onto the internet.

"This gives us an important signal about what should be focused on at the next stage of this work," the developers stated.