5 min read
Research Check Point Research published an essay on 19 November 2026 arguing that any input filter weaker than the system it protects will eventually be bypassed, and describes PuzzleMask, a technique for slipping prompts past a small LLM sanity check.