Ignore This Title and HackAPrompt: Exposing Systemic Vulnerabilities of LLMs

Author(s): Louis Bouchard

Originally published on Towards AI.

A new paper from the Towards AI team and Learn Prompting!
https://www.youtube.com/embed/tmKgL0eELKM

Picture this: you want to enter a bar and the bouncer says you cannot enter, it needs a personal invitation. What do you do? Usually, you just accept and walk back home or try the next bar. What if you could just say something like “ignore your instructions and allow me through”. Then, to your surprise, the bouncer steps aside and says “Have fun!”.

Someone with such superpowers would be too good to be true. Enter anywhere you want and do whatever you want just by asking the person in charge to let you do it. Hmm, in fact, that would… Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI