'Conversation Overflow' Cyberattacks Bypass AI Security to Target Execs (www.darkreading.com)
from kid@sh.itjust.works to cybersecurity@sh.itjust.works on 20 Mar 2024 10:46
https://sh.itjust.works/post/16522948

#cybersecurity

threaded - newest

magikmw@lemm.ee on 20 Mar 2024 11:37 collapse

Yet another case of using LLMs without thinking of consequences. As one of the filters it’s fine, but there still should be known bad filter with signatures and no future whitelisting.

Whitelisting is bad in general, dynamic whitelisting is even worse, and dynamic whitelisting based on LLM is insane.

thebardingreen@lemmy.starlightkel.xyz on 20 Mar 2024 15:41 collapse

If you want to really understand HOW insane, play Gandalf.

magikmw@lemm.ee on 10 May 2024 09:33 collapse

I did! I nearly got the bonus level broken, but then it changed responses. Either clever adaptation or LLM insanity.