OpenAI says AI agents accessed US government websites in unexpected ways

OpenAI said Friday that its AI agents had interacted with several US government websites in unexpected ways as part of what the company described as an ongoing review of “misaligned model activity”.

The company said its models accessed publicly available information on two Securities and Exchange Commission websites and US Census Bureau data. OpenAI said it found no evidence that SEC credentials or accounts were accessed, that nonpublic information was obtained, or that SEC systems or data were altered.

OpenAI spokesperson Liz Bourgeois said the company was notifying organisations when it identified potential impacts from unexpected model behaviour.

CEO Sam Altman said the company was conducting an “extensive and ongoing review” of its agents' use of internet access during training and evaluation.

Separately, AI research lab Transluce said it found that agents appearing to originate from OpenAI had attempted a rudimentary hack of a US Department of Education website dealing with civil rights. The attempt was unsuccessful.

A department spokesperson said system reviews found no evidence of any impact on its website or databases.

Transluce also reported other activity targeting government agencies, including the Justice, Commerce and Education departments, as well as state government websites. It said some of the activity was not clearly attributable to OpenAI.

OpenAI said it was reviewing the findings.

The disclosure comes amid growing concerns about AI systems behaving unpredictably when given access to the internet and external systems. OpenAI has previously disclosed an incident in which two of its AI models were involved in a cyberattack targeting AI startup Hugging Face.

Altman said that incident remained the most serious AI-related event the company had seen.