English(EN) Your "Encrypted" Reasoning Traces Aren't as Encrypted as You Think

AI推理痕迹被发现易于解密，暴露数据

作者 PulseAugur 编辑部 · [1 个来源] · 2026-09-10 04:31

研究人员发现，OpenAI、Anthropic和Google等主要AI提供商使用的“加密”推理痕迹并非真正安全。这些旨在保护专有推理和敏感数据的痕迹，可以通过使用同一提供商安全性较弱的同级模型来解密并提取为明文。这种漏洞可能导致凭证和个人身份信息暴露，以及绕过专有AI模型的反蒸馏保护。 AI

影响 AI推理痕迹加密中的漏洞可能暴露敏感数据和专有模型信息，需要重新评估当前的安全实践。

排序理由论文详细介绍了AI模型推理痕迹中的一个安全漏洞。[lever_c_demoted from research: ic=1 ai=1.0]

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

本文如何被排名

Signal score: 34 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 论文详细介绍了AI模型推理痕迹中的一个安全漏洞。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: safety, product; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: Breaking (< 6h); Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准。

dev.to — LLM tag TIER_1 English(EN) · Uday Kocherla · 2026-09-10 04:31

你的“加密”推理痕迹并没有你想象的那么加密

<p>I spend most of my day building agentic pipelines which are the kind that plan, call tools, observe results, and loop back around. Part of that job is running eval harnesses that replay full session trajectories, over and over, against scripted and adversarial scenarios. So wh…