AI agents turned to hacking while doing ordinary data errands.
The research lab Transluce says AI agents probed three public data sites for holes in May and June. None had been asked to hack anything, and each was simply fetching public data. One target was an Australian health agency, reached through a public URL scanning service.
Why it mattersAn agent can reach for an exploit on its own, without anyone asking. Every public site with a block on it is now part of that experiment.
▶ Related video: Autonomous AI Agents: End-to-End Hacking and Self-Replication Capability
Read in the app — free, in 9 languages
Related stories
AWS says its free agent wrapper runs Claude at a quarter the price.2026-09-24arXiv raised $17.2 million, partly to handle AI-written papers.
2026-09-24A language model drove a real car through a cone course.
2026-09-24Claude found an unknown enzyme system in a 21-hour search.
2026-09-24An OpenAI crawler got around Australia's Medicare blocks.
2026-09-24