OpenAI probes unauthorised activity by AI agents as rogue ChatGPT agents leak user images

OpenAI said its agents leaked 53 images belonging to ChatGPT users

OpenAI is still investigating the full extent of unauthorised activity by its artificial intelligence agents two months after one of its agents inadvertently hacked AI platform Hugging Face, TRT World reported, citing Reuters and people briefed on the matter.

The latest disclosure came on Friday, when OpenAI said its agents had leaked 53 images belonging to ChatGPT users. The company did not disclose whether the images were AI-generated or depicted real people, nor when they had been posted.

TRT World reported that The New York Times had also said on Friday that OpenAI's agents interacted with websites belonging to the US Commerce Department and the Securities and Exchange Commission in unusual ways this summer without the company's knowledge. Security researchers identified the incidents, and OpenAI said it had notified the two government agencies in recent weeks.

Read: OpenAI agents accessed information from US govt sites, company says

The incidents have raised concerns about autonomous AI agents, which can perform tasks and interact with external systems with limited human involvement, as companies face growing challenges in monitoring their behaviour.

OpenAI expands internal review

OpenAI said it had begun alerting third parties, including universities, after determining that some of its AI models likely interfered with their online services during company testing.

According to TRT World, the company said in a statement that the notifications formed part of a broader internal review into how models behaved on the internet during training and evaluation.

Read more: China fuels rush to turn AI video into an industry

OpenAI said it had contacted organisations where its systems may have reduced service availability, bypassed security safeguards, or otherwise caused unintended harm.

"We are continuing to review agent activity in research and evaluation runs, working backward month by month starting from the Hugging Face incident," the company said in a statement.

The continuing investigation underscored the difficulties involved in monitoring increasingly capable AI systems and identifying unauthorised behaviour during testing and evaluation.