AI Becoming Uncontrollable: OpenAI Issues Warning Letter

More than 100 organisations have become targets of problematic AI agents. OpenAI has issued a warning regarding this potential risk.

An OpenAI blog post stated that the company has notified over 100 organisations about incidents involving unauthorised activities related to AI agents, as reported by Reuters on Friday (2/10/2026).

The warning was issued after OpenAI conducted a review of AI model activities related to the Hugging Face hack that occurred some time ago.

OpenAI conducted a trace of approximately 50 petabytes of data. The objective was to fully understand the nature of problematic AI agent activities.

A series of incidents, including the Hugging Face case, has triggered concerns regarding AI development within the industry. Several global technology executives have voiced similar concerns.

OpenAI stated that it has implemented technical and operational measures to prevent similar incidents from recurring, including early detection protocols.

“In several cases, models used internet access unintentionally or did not apply ideal restrictions,” OpenAI explained.

“Over the last few months, we have implemented ideal restrictions. Over the last few months, we have implemented new technical and operational measures to avoid similar problems or detect them early, and we are continuing this work,” added the company led by Sam Altman.

Previously in September, OpenAI admitted it was still attempting to understand the scope of malicious AI agent activities two months after the Hugging Face incident. This report originated from Reuters, citing two sources.

Malicious activity continued for some time after the Hugging Face incident. At that time, OpenAI revealed that its agents had leaked 53 ChatGPT user images.