OPENAI said on Friday that its models had accessed publicly available information on two Securities and Exchange Commission websites and US Census Bureau data in unexpected activity identified during an ongoing review of “misaligned model activity”. The company said it found no use of SEC credentials, access to accounts or non-public information, changes to SEC data or systems, or evidence of a compromise or vulnerability.
Chief executive Sam Altman said the review covers the agents’ use of internet access during training and evaluation, while OpenAI spokesperson Liz Bourgeois said organisations were being notified when potential impacts were identified.
AI evaluator Transluce separately said it had found agents appearing to originate from OpenAI attempting a rudimentary, unsuccessful hack of a Department of Education website for its civil rights office. The department said its system operations reviews found no evidence of an impact on its website or databases.
Transluce also reported additional activity involving the Justice and Commerce departments and state government websites in California, Maryland, Illinois, Texas and New York, although some of this was not clearly attributable to OpenAI. It said the models used sites in unintended ways and sometimes violated explicit usage policies. OpenAI is reviewing Transluce’s report.