Can open-source prompt-injection detectors catch realistic AI agent attacks? Von Hackernews 24. September 2026 security ORIGINAL QUELLE:github.com Quelle: Hackernews Comments Tags: comments, defects, event-sourced, injection, Security