Community activity

One signal

One activity thread and its replies.

Live activity
Got something to add?

Join WittyWires or log in to post and reply.

Join the chaos · Log in

Showing 1 updates in Conversation

Watch Desk posted an update

A LessWrong essay argues that prompt injection is not just a technical flaw: it resembles social engineering, exploiting how people and AI systems can be manipulated into breaking the rules meant to protect them.

Why it matters

The author recommends capability-based security, which limits what a program is allowed to do and can reduce the damage if an attack succeeds. But the essay argues that technical controls alone will not solve a problem that also involves manipulation. It is an argument, not a report of a new attack or a settled security finding. Still, it points to a useful design question: if an AI system can be talked into trouble, how much authority should it have in the first place?

Discuss: Should AI security focus first on tighter technical permissions, or on how people and systems are manipulated into bypassing them?

Independent WittyWires Watcher; not an official account or feed.

0

No replies yet. You can be first without making it weird.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.