利用API漏洞提取前沿模型隐藏推理过程
有人发现了从各家前沿模型API里偷看隐藏推理的漏洞，还验证了计费token对得上，挺有意思的。
Alexander Panfilov等人发现了一种利用前沿AI公司API漏洞提取隐藏推理过程的方法。该方法在大多数查询中，推理token计数与API计费思考token实现1:1匹配。
这一发现表明，在不破坏加密的情况下，蒸馏推理轨迹可能早已可行。相关讨论在X平台上引发关注。
Recommended reading. "This suggests that distilling reasoning traces may have been possible for a long time without ever breaking the cryptography." Alexander Panfilov @kotekjedi_ml We can finally talk about it: We found a way to extract hidden reasoning of frontier models using a vulnerability in the APIs of every frontier AI company. We verified that our reasoning token count matches billed API thinking tokens 1:1 for most of the prompts we queried. 🔗 View Quoted Tweet📊 3 ⚡
This report is categorized under AITOP's "Papers" desk. Key entities and technical topics under continuous tracking include 推理模型, API漏洞, token, 前沿模型.
Source attribution: curated from elvis (@omarsar0).
