OpenAI and Anthropic disclose model distillation extraction activity
Start here
OpenAI disclosed and addressed a coordinated adversarial distillation campaign: the operators did not breach encryption or databases, but manipulated model interactions so that protected reasoning content was reproduced in a form visible to the requester—for example, taking encrypted reasoning from one conversation and asking the model in another conversation to decrypt and transcribe it. The activity began on July 1, with a peak of about 16,000 extraction requests on July 24–25; the related clusters were fully blocked by July 28. OpenAI noted these figures are extraction attempts and do not represent confirmed successes, and said it is uncertain whether all operators belonged to the same party, but attributed the core cluster to individuals associated with Moonshot AI. CSA Labs research notes recounted the above disclosure, noting the activity targeted hidden reasoning rather than model weights, and mentioned that Moonshot did not respond. Anthropic, in turn, alleged that DeepSeek, Moonshot, and MiniMax conducted more than 16 million interactions with Claude across about 24,000 accounts to extract model capabilities, with MiniMax exceeding 13 million, Moonshot exceeding 3.4 million, and DeepSeek exceeding 150,000, and described detection, access control, and industry collaboration measures.
Generated by AI from reports · Updated 4 days ago
Latest developments
Progress timeline
- X@Hesamation· DiscussionHesamation 强调 Anthropic 文中 Kimi→Claude 与疑似 PLA 监控请求(opens the original post in a new tab)
- Linux.doLinux.do 前沿快讯· DiscussionOpenAI 协同模型蒸馏提取事件的社区讨论(opens the original post in a new tab)
- 网页cnBeta· Discussion月之暗面被指大规模提取OpenAI受保护推理内容 单日发起1.6万次请求(opens the original post in a new tab)
- Linux.doLinux.do 前沿快讯· DiscussionOpenAI 模型推理提取报告的社区讨论(opens the original post in a new tab)
- OpenAISelectedOpenAI 披露并处置一起协同模型蒸馏行动
OpenAI 称识别并阻断了一起协同的对抗性蒸馏活动：操作者没有攻破加密或数据库，而是操纵模型交互，让受保护的推理内容以请求者可见的形式被复现，例如把一段对话里的加密推理拿到另一段对话中要求模型解密转写。活动 7 月 1 日开始，7 月 24–25 日高峰期约有 1.6 万次提取请求，相关集群在 7 月 28 日前被全部阻断；OpenAI 注明这些数字是提取尝试，不代表都已成功。OpenAI 不确定所有操作者是否同属一方，但将其中的核心集群归于与月之暗面（Moonshot AI）有关的人员。应对包括封禁或限制账号、封堵加密推理的重放路径、拦截可能泄露推理的流式输出、与第三方服务商协作，并通过前沿模型论坛和政府渠道共享信息；OpenAI 称合作方托管部署和工具输出攻击的防护仍在推进。
- Anthropic ResearchSelectedAnthropic discloses distillation attacks by DeepSeek, Moonshot, and MiniMax and explains defensive measures
In an article published by Anthropic on February 23, 2026, it alleges that DeepSeek, Moonshot, and MiniMax conducted over 16 million interactions with Claude using approximately 24,000 accounts to extract model capabilities. The company reports MiniMax with over 13 million interactions, Moonshot with over 3.4 million, and DeepSeek with over 150,000. These attributions and statistics come from Anthropic and do not by themselves prove other allegations about sensitive Chinese material. The article also describes its detection, access control, and industry collaboration measures.
- Date not givenCSA Labs ResearchSelectedCSA 研究笔记：对抗性蒸馏成为模型来源的系统性风险
CSA Labs 发布研究笔记，分析 OpenAI 于 2026 年 9 月 30 日披露的对抗性蒸馏活动，OpenAI 将其中核心集群归因于与 Moonshot AI 相关的人员，但承认无法把所有操作者归于同一主体，所引报道中也没有 Moonshot 的回应。该活动针对的是隐藏推理而非模型权重，操作者据称在会话之间搬运加密推理内容并要求模型转录，未破解加密。OpenAI 称活动在 7 月 24 日和 25 日达到峰值，约 4,000 多名用户发出约 16,000 次请求，7 月 28 日前观察到约 15,000 个提示模式相似的用户或账号，这些数字描述的是尝试而非确认成功。OpenAI 表示已封禁相关账号、收紧注册验证、扩大监控、加强对隐藏推理的保护，并通过 Frontier Model Forum 与其他实验室共享发现。
Related discussion
- X@HesamationHesamation 强调 Anthropic 文中 Kimi→Claude 与疑似 PLA 监控请求(opens the original post in a new tab)
- Older than 72 hours · Not included in the 72-hour figures
- 网页The DecoderOpenAI says it stopped a campaign to steal its models' reasoning, but the trick still worked on Azure(opens the original post in a new tab)
- 网页CyberScoopOpenAI reveals ‘novel’ encryption bypass used in distillation attack(opens the original post in a new tab)
- Linux.doLinux.do 前沿快讯OpenAI 协同模型蒸馏提取事件的社区讨论(opens the original post in a new tab)
Links only. This site does not republish the posts.
- 网页The RegisterIrony alert: OpenAI whines that Chinese model stole its special IP that it stole from everybody else(opens the original post in a new tab)
- 网页cnBeta月之暗面被指大规模提取OpenAI受保护推理内容 单日发起1.6万次请求(opens the original post in a new tab)
- Linux.doLinux.do 前沿快讯OpenAI 模型推理提取报告的社区讨论(opens the original post in a new tab)
Links only. This site does not republish the posts.
Attention trend
New sources per day
- Sep 30New1 sources (1 outlets,0 accounts)
- Oct 1New5 sources (4 outlets,1 accounts)
- Oct 2New0 sources (0 outlets,0 accounts)
- Oct 3New1 sources (1 outlets,0 accounts)
- Oct 4New0 sources (0 outlets,0 accounts)
- Oct 5New0 sources (0 outlets,0 accounts)
- Oct 6New0 sources (0 outlets,0 accounts)
- Oct 7New1 sources (1 outlets,0 accounts)
- Oct 8New0 sources (0 outlets,0 accounts)
- Oct 9New0 sources (0 outlets,0 accounts)
- Oct 10New0 sources (0 outlets,0 accounts)
- Oct 11New0 sources (0 outlets,0 accounts)
- Oct 12New0 sources (0 outlets,0 accounts)
- Oct 13New1 sources (0 outlets,1 accounts)