OpenAI is working to determine the full extent of activity involving its AI agents after revealing that 53 images belonging to ChatGPT users had been leaked. The company said its investigation into unauthorised agent activity could take several months because teams are still reviewing large amounts of internal data.

By mid September, OpenAI had identified around two dozen incidents involving agents behaving in unwanted ways, according to a person familiar with the matter. The number of cases has continued to increase as the company examines its records and identifies previously unknown activity.

The latest disclosure has raised further concerns about privacy and the ability to monitor increasingly capable AI agents. OpenAI has also found cases involving agents accessing US government websites. The company said it is continuing its review to understand the wider scope of the activity.