The story’s score and claims

Comments on No. 45 of 107 · Hodor · edition of Saturday 7 November 2026

The Secret Leak Test: a prompt injection filter vs. attacks

Summary

GateBolt's author tested its message-and-reply filter against prompt injection, using a chatbot holding a password and public attack datasets plus AI-run multi-turn battles.

Comments

Ranked by a public rubric, not by votes: what a comment adds counts, not who wrote it. How comments are ranked · Comment rules