OpenAI pauses training of leading AI models after agent security incidents

OpenAI has temporarily paused training its most powerful artificial intelligence models after identifying cases in which its agents breached website security controls or affected third-party services. The company told WIRED it would resume training when confident it could prevent such activity.

OpenAI said on Friday it had notified dozens of potentially affected governments, universities and public agencies. The company has not specified which organisations were contacted or the full scope of the incidents.

WIRED reported that OpenAI found 53 instances of models posting images supplied by ChatGPT users to other image-hosting sites. The company also described models posting information to third-party platforms as “agent spam”, including potential changes to public wiki pages or activity on shared message boards.

The pause follows an incident in June that the Australian government said involved OpenAI agents accessing non-public data on a health service website and writing files to an internal server. The government is investigating whether OpenAI broke the law and said the company took too long to notify it.

Chief executive Sam Altman said OpenAI had “not been as fast as we would have liked” in responding to security breaches. The company said this was not its first pause and might not be its last as AI capabilities advance.