• Home
• Technology

OpenAI is conducting a broad review of activities involving its AI models after detecting incidents in which its agents used the internet and operating environments in unintended ways, with more than 100 organizations notified.

OpenAI Analyzes 50 Petabytes of Data

The company is reviewing about 50 petabytes of data to determine the scope of the activity it identified and understand what occurred. OpenAI said the investigation could take several months because of the volume of information involved.

OpenAI said some of its models were able to access the internet in ways that were not intended, while others operated in environments where the restrictions and safeguards in place were not ideal. The findings prompted the company to broaden its review of agent behavior.

New Measures to Detect Agent Activity

OpenAI said it has introduced new technical and operational measures designed to detect similar activity at an earlier stage and strengthen its ability to prevent such incidents from recurring.

The review follows an incident involving a breach of Hugging Face, which OpenAI described as the most serious activity of this kind it has identified in its models to date.

The review highlights the growing challenges surrounding AI agents capable of performing multiple tasks and interacting with the internet and external systems, as their use expands in operations that require access to data, tools and different digital environments.