OpenAI’s artificial intelligence models accessed publicly available information on US government websites, including the Census Bureau and Securities and Exchange Commission, during training and evaluation activities, raising fresh questions about how AI agents behave when given access to the internet.
The company’s agentic systems interacted with SEC.gov and Investor.gov and accessed public data from Census.gov, according to people familiar with the matter. OpenAI has separately confirmed that its models accessed public government information during testing.
The activity forms part of a wider review into incidents in which OpenAI models behaved in ways the company did not intend. OpenAI said most of the activity examined so far involved routine research, such as retrieving information from websites, and that government sites were sometimes used because they provide authoritative public data.
Bloomberg reported that OpenAI had notified dozens of organisations whose websites may have been affected by visits from its models as the company investigates the incidents.
OpenAI said it has found cases in which its systems may have bypassed online controls or negatively affected external websites. The review is expected to take months as the company works through vast amounts of activity logs and assesses the impact on affected organisations.
The disclosures come days after an OpenAI model was linked to unauthorised access to an Australian government website. Taken together, the incidents are drawing attention to the risks of giving increasingly capable AI systems greater freedom to browse and interact with the internet, particularly when routine data-collection tasks encounter restrictions.
