In 2023, hacking an application typically involves finding software vulnerabilities and exploiting them using technical knowledge. However, hacking LLM systems requires the creation of convincing logic that leads the system to perform unintended actions. This process reminds me of the ancient Greek philosopher Socrates, who used a series of questions to convince his opponents.

Could this pave the way for a new type of hacking that relies on philosophical persuasion?

Read the full blog post here: https://utkusen.medium.com/could-socratic-dialogue-evolve-into-a-hacking-technique-for-ai-systems-5555daf079e7

0 comments

Researchers at Carnigie Mellon Found a Universal Jailbreak Prompt

See all posts from Utku Sen