安全★★★★Simon Willison · 2026-07-25
Boris Cherny on Opus 5's Resistance to Prompt Injection
Boris Cherny highlights that Opus 5 is their least prompt-injectable model yet, emphasizing its strong resistance in PI evaluations and red teaming.
📌 Key points
- Opus 5 demonstrates exceptional resistance to prompt injection in PI evaluations
- This finding is more exciting than any evaluation score
- Prompt injection is a critical issue in AI security
- The discovery has significant implications for enhancing AI system safety
本页为 gitzw.com 基于公开来源的 AI 中文解读,非原文转载。