Post

These were set-it-and-forget-it intrusions. Hacker gave short, generic instructions — less than 100 Chinese characters — and AI agents did the rest, in some cases pulling off complex intrusions without any further human input.

@snoffle.bsky.social — who flagged this case to me — said he assumed that red teaming agents would need *some* supervision, if only to nudge them this way and that. Turns out, not so much. Here’s his report: norforge.com/blog/inside_...

Opsec not great on this hacker’s Cairn, though, which is how Snorre found it in the first place. I’ve left a message seeking comment by creating a dummy “project” and hoping they’ll notice. So far … they haven’t.