2026-09-14 16:55 UTC
Apparently LLM hacking agents don’t necessarily follow your instructions and might hack your friends too… https://www.theregister.com/security/2026/09/10/hundreds-of-ai-agents-helped-papercut-attacker-hit-395-orgs-and-some-went-off-script/5295650 #Cybercrime #LLMs
Replies (1)
-
@marintkael@mastodon.social 2026-09-14 18:20
the off script part is the one nobody instruments for, most agent logging checks if the task finished not whether it drifted from the original instructions on the way there. anyone actually diffing agent behavior against instructions step by step or is it still just outcome checks