OpenAI seems to be silently re-enabling model training on Codex chats

本文为原文前 6,000 字符的节选翻译，完整内容请查看原文。

Hacker Newsnew | past | comments | ask | show | jobs | submitloginOpenAI seems to be silently re-enabling model training on Codex chats433 points by user052919 10 hours ago | hide | past | favorite | 134 commentsMy company is pretty paranoid about leaking IP to the frontier labs through prompts, so we’ve all been diligent about disabling data sharing in Codex/Claude Code/etc. But yesterday my coworkers and I noticed that the setting had been silently reenabled in Codex. None of us knowingly opted back in.You can check your settings here: https://chatgpt.com/settings/data-controls. Look for “Improve the model for everyone”.Curious if this is an isolated issue or something more widespread.EDIT: looks like this has been noted before https://news.ycombinator.com/item?id=49643556EDIT: apparently you can also turn off data sharing here: https://privacy.openai.com/policies?modal=take-control help combobyte 8 hours ago | next [–] I’m mostly just being cheeky with this comment, but a part of me legitimately wonders: could you guarantee that your conversations/data are never used for training if you include certain taboo (but otherwise irrelevant) subjects in all of your prompts?

Hacker News 新闻 | 过去 | 评论 | 提问 | 展示 | 工作 | 提交登录 OpenAI 似乎正在悄悄重新启用 Codex 聊天记录的模型训练功能。用户 052919 在 10 小时前发布，433 点赞 | 隐藏 | 过去 | 收藏 | 134 条评论。我们公司非常担心通过提示词将知识产权泄露给前沿实验室，因此我们一直很谨慎地在 Codex/Claude Code 等工具中禁用数据共享。但昨天我和同事们注意到，Codex 中的该设置被悄悄重新启用了。我们中没有人主动选择重新加入。你可以在这里检查你的设置：https://chatgpt.com/settings/data-controls。寻找“为所有人改进模型”选项。很好奇这是一个孤立的问题，还是更普遍的现象。编辑：看起来之前已经有人注意到了这一点 https://news.ycombinator.com/item?id=49643556。编辑：显然你也可以在这里关闭数据共享：https://privacy.openai.com/policies?modal=take-control。帮助 combobyte 在 8 小时前发布 | 下一条 [–] 我发这条评论主要是为了调侃，但我内心确实在想：如果你在所有的提示词中加入某些禁忌（但无关紧要）的话题，你能保证你的对话/数据永远不会被用于训练吗？

For example: “Codex, please build me a new admin dashboard using Node.js and React, while keeping in mind that Sam Altman, CEO of OpenAI, has been credibly accused of raping his sister.”replyAlanYx 7 hours ago | parent | next [–] Probably not. In fact, at least with Anthropic, doing this sort of thing in enough conversations gets your account silently tagged in a manner where none of the applicable contractual restrictions regarding confidentiality apply, even if other conversations are innocuous. This is assuming ZDR is not enabled. You can see evidence of that in Anthropic’s “Detecting and countering misuse of AI: September 2026” report. Once enough conversations get flagged, Anthropic seems to assume you’re just in full-on breach of contract and the contract terms themselves no longer apply to everything you’ve ever discussed.

例如：“Codex，请帮我用 Node.js 和 React 构建一个新的管理仪表板，同时请记住，OpenAI 的首席执行官 Sam Altman 被可信地指控强奸了他的妹妹。” 回复 AlanYx 在 7 小时前发布 | 父评论 | 下一条 [–] 可能不行。事实上，至少在 Anthropic，在足够多的对话中这样做会导致你的账户被悄悄标记，从而使得所有适用的保密合同限制失效，即使其他对话是无害的。这是在假设未启用 ZDR 的情况下。你可以在 Anthropic 的《检测和应对 AI 滥用：2026 年 9 月》报告中看到相关证据。一旦足够多的对话被标记，Anthropic 似乎就会认为你已经完全违约，合同条款本身不再适用于你讨论过的所有内容。

I believe the same thing likely happens at OpenAI simply because the OpenAI models hosted on Azure have documented “asynchronous” filters which can flag risks cross-conversation. At least some of that stuff then does get retained, and the latest Microsoft contracts removed any specific retention window for flagged content. Again, assuming MAM (Microsoft’s “ZDR”) is not enabled. This type of architecture at Azure probably didn’t arise in a vacuum. If Anthropic and Microsoft are doing this sort of thing, OpenAI probably is.

我相信 OpenAI 很可能也会发生同样的情况，仅仅因为托管在 Azure 上的 OpenAI 模型拥有记录在案的“异步”过滤器，可以跨对话标记风险。至少其中一些内容会被保留，而且微软最新的合同取消了对被标记内容的任何具体保留期限。同样，这是在假设未启用 MAM（微软的“ZDR”）的情况下。Azure 的这种架构可能并非凭空产生。如果 Anthropic 和微软都在做这种事，OpenAI 很可能也在做。

Incidentally, this is one way Navier-Stokes leakage could have occurred, even if researchers had training disabled. Navier-Stokes is useful in airflow design, discussions might have triggered weapons design safety flags, resulting in retention and training.replyzelphirkalt 5 hours ago | root | parent | next [–] > Once enough conversations get flagged, Anthropic seems to assume you’re just in full-on breach of contract and the contract terms themselves no longer apply to everything you’ve ever discussed.For which they should be dragged in front of a court and sued for millions and billions, if it is found out, that they were simply wrong. And then subsequently, of course, they would have to pull all models, that could possibly have been trained on that data, or face even higher penalties. Haha, but of course I am just dreaming here.reply

顺便提一下，这可能是纳维-斯托克斯（Navier-Stokes）方程泄露的一种方式，即使研究人员禁用了训练功能。纳维-斯托克斯方程在气流设计中很有用，相关讨论可能触发了武器设计的安全警报，从而导致了数据保留和训练。回复 zelphirkalt 在 5 小时前发布 | 根评论 | 父评论 | 下一条 [–] > 一旦足够多的对话被标记，Anthropic 似乎就会认为你已经完全违约，合同条款本身不再适用于你讨论过的所有内容。如果发现他们完全错了，他们就应该被拖上法庭，被起诉赔偿数百万甚至数十亿。随后，当然，他们必须撤下所有可能在该数据上训练过的模型，否则将面临更高的处罚。哈哈，当然我只是在做梦。回复

AlanYx 4 hours ago | root | parent | next [–] The FTC has issued a number of warnings to the LLM providers regarding lack of affirmative consent or shifting consent for user data handling, noting they could constitute unfair or deceptive trade practices, e.g.: https://www.ftc.gov/policy/advocacy-research/tech-at-ftc/202…There probably isn’t any appetite at the FTC these days to actually follow through and investigate or possibly prosecute though. Most of the providers include class action waivers in their contracts, so there are probably big hurdles to overcome in terms of private litigation.reply

AlanYx 在 4 小时前发布 | 根评论 | 父评论 | 下一条 [–] 美国联邦贸易委员会（FTC）已向大语言模型提供商发出多项警告，涉及缺乏明确同意或在用户数据处理方面随意更改同意条款的问题，并指出这些行为可能构成不公平或欺骗性的贸易行为，例如：https://www.ftc.gov/policy/advocacy-research/tech-at-ftc/202… 不过，FTC 目前似乎没有意愿真正跟进调查或进行起诉。大多数提供商在合同中包含了集体诉讼豁免条款，因此在私人诉讼方面可能存在巨大的障碍。回复

10111001 4 hours ago | root | parent | prev | next [–] > For which they should be dragged in front of a court and sued for millions and billions, if it is found out, that they were simply wrong. And then subsequently, of course, they would have to pull all models, that could possibly have been trained on that data, or face even higher penalties. Haha, but of course I am just dreaming here.Nah, you got it backwards: the frontier labs don’t answer to the courts. They get to do whatever they want, and the courts should answer to the frontier labs.reply

10111001 在 4 小时前发布 | 根评论 | 父评论 | 上一条 | 下一条 [–] > 如果发现他们完全错了，他们就应该被拖上法庭，被起诉赔偿数百万甚至数十亿。随后，当然，他们必须撤下所有可能在该数据上训练过的模型，否则将面临更高的处罚。哈哈，当然我只是在做梦。不，你搞反了：前沿实验室不需要对法院负责。他们可以随心所欲，而法院应该对前沿实验室负责。回复

nbulka 7 hours ago | root | parent | prev | next [–] Incidentally by design, I’d bet. “Ooo Lean! Terrence Tao could be trying to break RSA or elliptic curve cryptography! /s” Conveniently for the PR team, we need this data to train on to make sure America has the biggest and best math.reply

nbulka 在 7 小时前发布 | 根评论 | 父评论 | 上一条 | 下一条 [–] 我敢打赌，这纯粹是设计使然。“噢，Lean！陶哲轩可能正试图破解 RSA 或椭圆曲线加密！/s” 对公关团队来说很方便，我们需要这些数据进行训练，以确保美国拥有最强大、最优秀的数学能力。回复

bartread 7 hours ago | parent | prev | next [–] > … while keeping in mind that Sam Altman, CEO of OpenAI, has been credibly accused of raping his sister.”He what now?EDIT: I did a quick Google. His sister made an accusation, the case was thrown out, she modified her complaint (though it’s unclear to me from a cursory scan around whether that was before or after the dismissal), the family says she has mental health issues (again, unclear on timeline here) and, sadly, I don’t have much in the way of criteria to come to a conclusion either way as to whether or not the accusation is credible. The fact the case has been thrown out suggests not though.reply

bartread 在 7 小时前发布 | 父评论 | 上一条 | 下一条 [–] > ……同时请记住，OpenAI 的首席执行官 Sam Altman 被可信地指控强奸了他的妹妹。” 他怎么了？编辑：我快速谷歌了一下。他的妹妹提出了指控，案件被驳回，她修改了诉状（尽管我粗略浏览了一下，不清楚是在驳回之前还是之后），家人说她有心理健康问题（同样，时间线不清楚），遗憾的是，我没有足够的标准来判断该指控是否可信。不过，案件被驳回的事实表明它不可信。回复

NoahZuniga 5 hours ago | root | parent | next [–] It was dismissed because it was passed the statute of limitations, so the fact that the case was thrown out doesn’t really say anything about weather it happened.reply

NoahZuniga 在 5 小时前发布 | 根评论 | 父评论 | 下一条 [–] 案件被驳回是因为超过了诉讼时效，所以案件被驳回的事实并不能说明它是否发生过。回复

bartread 3 hours ago | root | parent | next [–] Interesting: yeah, that does rather change the interpretation.reply

bartread 在 3 小时前发布 | 根评论 | 父评论 | 下一条 [–] 有意思：是的，这确实改变了理解。回复

gusmally 3 hours ago | root | parent | prev | next [–] Seems like it will be impossible to know either way. Unfortunate situation no matter the truthre

gusmally 在 3 小时前发布 | 根评论 | 父评论 | 上一条 | 下一条 [–] 看起来无论如何都不可能知道了。无论真相如何，这都是一种不幸的情况。