Charles R. Twardy

Follow @ctwardy on Micro.blog.

Fooling AI

Cyber-security is a broken-window fallacy, but there’s something delightful about this little bot tarpit:

The attacking bot reads the hidden prompt and often traverses the infinite tarpit looking for the good stuff. From Prompt Injection as a Defense Against LLM-drive Cyberattacks (two GMU authors!). HTT Unsupervised Learning (Daniel Miessler)