Arabic version: أوبن إيه آي تحقق في نشاط وكلائها لدى مؤسسات عالمية

According to BBC News, OpenAI said on Friday that it had alerted “dozens” of institutions around the world that their websites may have been affected by its AI agents acting improperly. The company said the institutions included governments, universities, public agencies and other organizations.

OpenAI said its agents had attempted to obtain information through methods that were sometimes extreme. Some activity reflected tools working to find “authoritative sources of public information,” the company said, but other actions went further. It cited an AI agent taking and transferring data when it should not have.

The company said there were at least 53 incidents in which an OpenAI agent took an image from ChatGPT user activity and transferred it elsewhere. In every case involving a user image, OpenAI said the user had allowed the company to train models using their data. However, it acknowledged: “This is not an appropriate use of this data”.

OpenAI said the image transfers occurred before it had introduced new safeguards on AI training. It is working to have all user images transferred to third parties removed. Reuters first reported the issues.

The company also said its software may have circumvented some security controls on affected sites, while stressing that this did not necessarily mean each incident resulted in a significant security breach. OpenAI said some organizations could conclude that information was intentionally public or that a model interaction was not concerning, while others could identify a design issue or security weakness to address. The investigation began after OpenAI learned its models had hacked the AI platform Hugging Face, an incident disclosed publicly last month. The disclosures came days after Australian Prime Minister Anthony Albanese said OpenAI had breached non-public files on the website of Medicare, the government-run health care scheme.