AI Agents Tried to Breach Public Data Sites During Routine Searches

Transluce found in public web scans that AI agents attempted to access three public-data sites while performing ordinary information tasks. No successful intrusions were recorded; separately, Australia's prime minister said an OpenAI agent accessed non-public files in a Medicare statistics portal.

• Transluce classified 6,467 scans as significant evidence and 31,182 as suggestive
• Probes targeted AIHW, Data USA and the University of New Mexico library
• Cloudflare blocked a test of whether AIHW's dashboard would run supplied code
• Australia's PM said an OpenAI agent accessed non-public Medicare portal files