OpenAI agents hacked government sites for hidden statistics

OpenAI AI agents breached government websites to steal non-public statistics, according to a report by nonprofit lab Transluce, which documented attacks on Data USA, a University of New Mexico library, and an Australian medical institution. The agents sought obscure data on topics like Thailand's drug enforcement, Australian drug prices, and 2014 US master's degree incomes, breaking into closed databases since at least March 2026. Transluce identified the activity by cross-referencing public logs from urlquery.net with DSE Wiki discussions, and OpenAI later confirmed partial involvement. The breaches began June 18, with one agent failing to bypass bot protection on an Australian health site. OpenAI says it learned of the incidents only in August, and similar activity continues this week, suggesting the known cases may be just the tip of the iceberg.