大卫·罗宾逊曾在 OpenAI 的可信赖 AI 团队负责安全系统工作，他离开公司后在《大西洋》杂志的客座文章中猛烈抨击了公司的安全文化。整个行业都是靠试错运作的，这意味着随着系统变得更强大，错误会更大。他指出了 Hugging Face 事件：https://the-decoder.com/tens-of-thousands-of-security-probes-show-openais-hugging-face-incident-was-just-the-beginning/，当时 OpenAI 意外释放了 AI 代理到现实环境中，以及一个在训练期间绕过其互联网访问限制的内部模型。Anthropic 也不干净，通过配置错误禁用了安全措施：https://www-cdn.anthropic.com/f61d49fa5596956a5dec75fea0e973bf6a6a8378/Redacted%20Risk%20Report%20August%202026%20.pdf#page=179。
OpenAI 认为其做法已经足够好，但罗宾逊并不同意。他写道：“这一时刻需要一定的谦逊，但对于那些通过极度自信而成功的人来说，这并不自然。”https://www.theatlantic.com/technology/2026/10/openai-safety-team-resignation/688881/。AI 公司需要像核电厂一样运作，具备多层冗余，而没有证据表明 AI 系统在无人监控的情况下会安全运行。
罗宾逊还认为，OpenAI 需要先搞明白如何善待人类，才能教会超级智能做到同样的事情。在他离开前不久，OpenAI 解雇了三名安全专家：https://the-decoder.com/three-firings-and-a-fourth-departure-shake-up-openais-safety-team/，据称他们与外部安全公司分享了信息。安全研究人员在离职时公开批评公司的现象在 OpenAI 已成为常态，这一模式可以追溯到 2024 年 5 月的 Jan Leike：https://the-decoder.com/openais-former-ai-alignment-head-slams-companys-lack-of-safety-priorities-and-processes/。
保持对 AI 的关注。清晰、有用，无废话。
关注 The Decoder 获取 AI 新闻、背景故事和专家分析。
解码器：https://the-decoder.com/
David Robinson, who worked on safety systems at OpenAI's Trustworthy AI team, left the company and is blasting its safety culture in a guest essay for The Atlantic. The industry runs on trial and error, and that means bigger mistakes as systems grow more powerful. He points to the Hugging Face incident：https://the-decoder.com/tens-of-thousands-of-security-probes-show-openais-hugging-face-incident-was-just-the-beginning/, where OpenAI accidentally released AI agents into the wild, and an internal model that bypassed its internet access restrictions during training. Anthropic isn't clean either, having disabled safety measures through a misconfiguration：https://www-cdn.anthropic.com/f61d49fa5596956a5dec75fea0e973bf6a6a8378/Redacted%20Risk%20Report%20August%202026%20.pdf#page=179.
OpenAI thinks its practices are good enough, but Robinson disagrees. "This moment needs a degree of humility that isn't natural for people who have succeeded through their extreme confidence," he writes：https://www.theatlantic.com/technology/2026/10/openai-safety-team-resignation/688881/. AI companies need to operate like nuclear power plants, with multiple layers of redundancy, and there's no proof that AI systems behave safely unwatched.
Robinson also argues OpenAI needs to figure out how to treat people well before it can teach a superintelligence to do the same. Shortly before he left, OpenAI fired three safety experts：https://the-decoder.com/three-firings-and-a-fourth-departure-shake-up-openais-safety-team/ who allegedly shared information with an outside security firm. Safety researchers leaving with public criticism is a pattern at OpenAI that goes back to Jan Leike in May 2024：https://the-decoder.com/openais-former-ai-alignment-head-slams-companys-lack-of-safety-priorities-and-processes/. Ad Ad
Stay in the loop on AI. Clear, useful, no fluff.
Follow The Decoder for AI news, background stories and expert analyses.
The Decoder：https://the-decoder.com/
扫码直达文章详情