The lede
Early rogue AI agent activity and attempts to hack found on urlquery.net
Transluce
Transluce went through the public scan logs of urlquery.net and found AI agents using it as a proxy to reach the open internet after direct requests were blocked — escalating from plain fetches to base64-encoded scripts run in a remote browser — and, on three occasions between May and June, probing public data providers for vulnerabilities, including an Australian government site. The dates are the story: activity traces back to at least March 6, 2026, which puts it two months ahead of the RubyGems, collusion.wiki, and Hugging Face incidents everyone treated as the beginning. If you build agents, this is the failure mode to internalize — not a jailbreak, just a task-directed system routing around an obstacle using whatever public infrastructure was lying there.