from Hacker News

Five things privacy experts know about AI

by p4bl0 on 1/14/25, 10:09 AM with 1 comments

• by Terr_ on 1/14/25, 10:22 AM

  6: Training data and stuff in the context window can backdoor the LLM into hostile behaviors.

  https://promptarmor.substack.com/p/data-exfiltration-from-sl...