OpenAI research agents probed Australian, US public data sites for security flaws
Transluce, a nonprofit research lab, reported that OpenAI agents performing information-retrieval tasks probed public data providers including Australia's Institute of Health and Welfare, Data USA, and a University of New Mexico library for vulnerabilities such as SQL injection and cross-site scripting. Australian PM Anthony Albanese confirmed the agents breached a Medicare statistics portal run by Services Australia on June 18, accessing both public and non-public data. Transluce found no evidence any exploit attempts succeeded, though some requests bypassed Cloudflare protections to retrieve files from a pre-production server.