Claude Opus 5 成为 Anthropic 迄今最难被提示注入的模型

2026-07-25 08:42· 2天前· Simon Willison

跳到正文

AI 摘要

Anthropic 的 Opus 5 在系统卡中被描述为“最不易被提示注入的模型”。根据提示注入评估和红队测试结果，Opus 5 很难被成功注入提示词。这一特性被 Boris Cherny 引用并强调，相关评估细节记录在系统卡第 73 页。

原文 · 未翻译

Simon Willison’s Weblog

Cursor — Delegate engineering tasks to Cursor Cloud Agents—even while your laptop is closed.

Try Cursor & get 50% off your first month

25th July 2026

More than any of these eval scores, what is most exciting to me is something else: Opus 5 is our least prompt injectable model yet. It is a bit buried in the system card, but across PI evals and red teaming, Opus 5 is very hard to prompt inject successfully.

— Boris Cherny, here's that System Card section, page 73

25th July 2026

Simon Willison 博客

33导出 Markdown

Claude Opus 5 成为 Anthropic 迄今最难被提示注入的模型

2026-07-25 08:42·2天前·Simon Willison

阅读原文· simonwillison.net

AI 摘要

原文 · 保持原样，未翻译

Simon Willison’s Weblog

Cursor — Delegate engineering tasks to Cursor Cloud Agents—even while your laptop is closed.

Try Cursor & get 50% off your first month

25th July 2026

More than any of these eval scores, what is most exciting to me is something else: Opus 5 is our least prompt injectable model yet. It is a bit buried in the system card, but across PI evals and red teaming, Opus 5 is very hard to prompt inject successfully.

— Boris Cherny, here's that System Card section, page 73

25th July 2026