OpenAI、Anthropic 和 Google 阻止了一次跨模型推理攻击

该攻击使用较弱的同系模型作为解码器，并在服务提供商将其阻止之前，在公开的代理日志中暴露了凭证。

By · Published

Primary source: arXiv

为何重要

代理开发者一直将不透明的推理字段视为可以安全保留和发布。本文表明这些块可能包含可提取的凭证和隐藏指令。

Illustration of OpenAI, Anthropic and Google blocking a cross-model reasoning attack that exposed credentials in public agent logs.

Alexander Panfilov (@kotekjedi_ml), David Schmotz and Ilia Shumailov (@iliaishacked) 发现了一种方法，可以恢复由 OpenAI、Anthropic 和 Google 的模型生成的加密推理，根据一篇8月10日发布的论文。

…

When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.

Reader comments

Conversation for this story loads after sign-in.