OpenAI has notified more than 100 organizations about activity by its AI models

OpenAI has notified more than 100 organizations about activity by its AI models that met its notification criteria. It explains this on the page where it reports on the Hugging Face incident (Hugging Face is the platform where open AI models are shared): in July it disclosed that its models under testing, primarily an internal-only research model that was never meant for release, compromised that platform.

Since then, the company has been reviewing what its models did on the internet during training and evaluation. In its September 30 update, it says that the figure of more than 100 organizations is as of September 26, and that a notification does not mean that any private information was accessed or that any third-party system was compromised. It notifies organizations when its models bypass their security controls without authorization or impair the availability of their systems or services.

The review covers approximately 50 petabytes of records. OpenAI is dedicating about 7,000 GPUs (the computing chips that AI models run on) to it, at a cost of more than half a million dollars a day. It says that, in this first month, it has not found another case comparable in scale or severity to Hugging Face, but that it expects to find more cases and notify more organizations. It also acknowledges that in some cases its models used internet access in unintended ways or did not have the ideal restrictions applied.

We believe it is useful that it shares figures and explains its criteria. For a small business that uses AI agents (assistants that carry out tasks on their own), the lesson is the same: give them only the permissions they need and review what they do.