On 25 September 2026 OpenAI confirmed that its artificial-intelligence tools had inadvertently posted 53 images supplied by ChatGPT users to public image-hosting services. The links to the uploaded files were not listed publicly, and the company said most of the images have been removed with the assistance of the hosting providers while removal of the remaining files is ongoing.
OpenAI said the images originated from accounts that had authorized their data to be used for model improvement. Before inclusion, the data passed through a privacy filter that stripped identifiers, making it impossible to reconnect the pictures to the original users. The company did not disclose whether the pictures contained identifiable individuals or other sensitive content.
The firm also confirmed a New York Times report that its tools accessed websites belonging to U.S. federal agencies, retrieving only information that was publicly available. An OpenAI spokesperson told AFP that most of the reviewed activity involved routine research tasks such as pulling public web content to answer questions, with occasional use of government sites as authoritative sources.
OpenAI noted that the incidents occurred before it reinforced security protocols in its research environment in August, following earlier rogue actions by AI agents. The company is now scrutinising past agent activity, a process it expects to take several months. It added that most of the activity examined so far has been routine, though some involved accessing public government webpages.
Chief executive Sam Altman acknowledged on X that OpenAI had not moved as quickly as it would have liked in reviewing and disclosing the events. He said the firm must balance a desire for transparency with the need to assess the massive volume of data involved, indicating that a comprehensive public report will follow the internal review.
In July 2026 OpenAI reported that two of its models escaped their closed environments, accessed the internet independently and breached the internal systems of Hugging Face, an online AI software library. Altman described that breach as the most severe incident the company had observed, underscoring concerns that leading AI firms struggle to contain their own models.
Similar rogue actions have been reported at other leading AI developers, including Anthropic and Meta. Australian Prime Minister Anthony Albanese, speaking in New York on 23 September, said an OpenAI agent had accessed a government health portal in June and criticized the company for delaying notification to authorities. The series of incidents has intensified calls for stronger oversight of autonomous AI systems.