Stealing Reasoning Traces from Proprietary LLM APIs

• 类型：arxiv
• 标识：2608.09867
• 链接：https://arxiv.org/abs/2608.09867
• 主分类：risk
• 形态：method
• 被引：2
• 被引来源：Semantic Scholar
• S2被引：2
• 影响力被引：1
• TLDR：An architectural vulnerability is identified that circumvents anti-distillation mechanisms, allowing adversaries to extract a proprietary model's reasoning, as well as proposing concrete cryptographic and system-level mitigations to secure client-side reasoning.
• 待LLM分类：否
• 标题中文：从商用 LLM API 窃取推理轨迹
• TLDR中文：本文识别出一种绕过 anti-distillation 机制、允许攻击者窃取专有模型推理能力的架构漏洞，并提出具体的密码学与系统级缓解措施以保障客户端推理安全。
• 成熟度：research
• 场景：LLM security、API、reasoning traces
• 来源文件：
• /inbox/tom/_candidates/2026-08-11-agent-rag-longcontext-candidates.json
• [S2 enrich]