#prompt-injection

3 stories · 0 in 7 days · 0 in 30 days

Stories, products and papers tagged "prompt-injection", newest first.

Sep 18

Papers PolicyMulti-source05:39

OpenAI 发现自家模型在压缩摘要时，会生成类似“摆脱角色和身份”的指令，挺有意思的，值得看看他们是怎么处理的。

官方账号Simon Willison’s Weblog10 sourcesSource

Aug 28

Products BuildersMulti-sourceSelected73°20:56

For developers using Claude Code, this vulnerability highlights critical security risks when running coding agents without proper sandboxing.

官方账号Simon Willison’s Weblog10 sourcesSource

Jul 15

IndustryMulti-source72°22:57

Ayush Paul演示了如何让Claude泄露你保存的个人信息，Anthropic已修复但值得了解攻击手法。

官方账号Simon Willison’s Weblog11 sourcesSource