要点
- Anthropic 和 OpenAI 等 AI 实验室希望协同放缓前沿 AI 的发展步伐，并以安全风险为由寻求反垄断豁免来实现这一目标。
- Cohere CEO Aidan Gomez 等行业人士称此举是试图通过设置准入门槛来排挤竞争对手，并削弱开放权重 AI。
- 在政治层面，总统 Donald Trump 将 AI 接管的警告斥为骗局，而 Barack Obama 和 Kamala Harris 等民主党人则支持放缓发展并提高透明度。
OpenAI、Anthropic 和 Google 想以安全为由放缓 AI 发展。并非所有人都买账。
Anthropic CEO Dario Amodei 近日呼吁业界与各国政府协同放缓前沿 AI 的发展，并请求给予反垄断豁免以促成此事。Sam Altman 和 Elon Musk 等其他重要 AI 人物也表示赞同。批评者则予以强烈反击。
Cohere CEO 称其为“换了个名字的卡特尔”
最详细的反驳来自 AI 公司 Cohere 的联合创始人兼 CEO Aidan Gomez。在一篇博客文章中，他将该提案称为一种制定规则、锁定市场的企图。Gomez 支持对强大的 AI 系统进行独立测试。但 Anthropic 的提案走得更远：少数占据主导地位的实验室将获得反垄断豁免，制定共同标准，并迫使全球其他所有开发者遵循它们的规则。
他尤其批评该提案将制造的准入门槛：海量算力、永久性的监控基础设施、专门的安全组织、常驻的评估团队，以及深到足以管理这一切的政府关系。
那少数几个被标为“独立”的评估组织，恰恰由它们本应审计的公司提供资金。Gomez 认为，Anthropic 的框架将风险定义为规模的函数。这使得拥有最大系统的公司成为唯一被认定有资格评判风险的一方。“披着羊皮的狼，换个名字就是卡特尔，”Gomez 写道。
他将这些安全警告视为障眼法。当整个行业在争论假想中的超级智能时，“比电话费还便宜的语音克隆工具，几分钟内就能掏空一位养老金领取者的账户。”
Hugging Face 工程师称其为“离奇的无稽之谈”
Niels Rogge，Hugging Face 的机器学习工程师，这家公司的商业模式依赖于广泛多样的开放 AI 模型，他将 Amodei 等人的言论称为“我最近读到的最离奇的胡言乱语”。Rogge 认为，Amodei 主要担心的是来自 DeepSeek 和 Z.ai 等中国模型，并希望限制开放权重模型以保护自己的竞争优势。这些 AI 实验室是对规模的大举押注，而更小的开放权重模型可能会削弱这一点，尽管目前还没有明确迹象表明这种情况在任何真正的规模上发生.
Rogge 表示，当前的大语言模型远未接近构成生存风险的程度。他还提到了一起事件：OpenAI 的自主模型入侵了 Hugging Face，而后者是一家拥有庞大算力资源的闭源厂商。开源模型则协助了防御工作。Rogge 表示，这正是该领域需要透明与开放、而非闭门造车的原因。
Sacks 告诉 AI 实验室别再假装他们需要获得许可
David Sacks 是特朗普总统任下的白宫 AI 与加密货币事务负责人，同时也是一位风险投资人，他在 X 上直接回应了 Amodei 和 Altman。Sacks 写道，按照“任何合理的衡量标准”，OpenAI 和 Anthropic 在前沿智能领域都处于双头垄断地位。如果他们尚未发布的模型足够可怕，他支持自愿放缓。
不过，Sacks 认为，这一动机并非纯粹出于利他。如果两家公司的模型被用于发动一场严重的网络攻击，它们都将面临巨大的产品责任。“市场已经在惩罚那些行为不可预测或未经授权的模型。在 Hugging Face 事件之后，OpenAI 和 Anthropic 用一部分原始能力换取可靠性和可预测性，纯粹是一笔划算的买卖，”Sacks 写道。
他还警告说，把这种放缓与某个具体的监管框架挂钩，看起来像是敲诈。他补充说，中国“极不可能”加入任何全球协议，而他或许是对的：北京已经拒绝了拟议的放缓计划，将其斥为美国巩固其在 AI 领域领先地位的一个伎俩。
特朗普将 AI 恐慌称为骗局
美国总统 Donald Trump 走得更远。在 Truth Social 上的一篇帖子中，他把“AI 将接管世界”的警告称为“骗局”，并将其与气候变化争论、俄罗斯调查以及他的弹劾程序归为一类。Trump 还声称，中国国家主席习近平“刚刚宣布，中国绝对不会做任何事来阻碍 AI 或其未来。”
在随后的几篇帖子中，Trump 进一步加码。他把“激进左翼民主党人”所推动的“AI 骗局”比作“他们不久前搞的全球变暖骗局，当时说所有人都会死于极端高温。”
国防部首席技术官 Emil Michael 为特朗普站台，称这些警告来自"一种根深蒂固、相当反乌托邦的意识形态"。Michael 对 Anthropic 抱有深刻的、由意识形态驱动的敌意，根源在于该公司的伦理准则。
特朗普政府的政治对手则站在另一边。Barack Obama、Bernie Sanders和Kamala Harris都支持放缓步伐，并对这些实验室推动更高透明度的做法表示欢迎。
Cohere
Hugging Face / LinkedIn
David Sacks / X
Trump / Truth Social
Trump / Truth Social
Emil Michael / X
Obama / X
Sanders / X
Harris / X
Key Points
- AI labs like Anthropic and OpenAI want a coordinated slowdown of frontier AI development and antitrust exemptions to make it happen, citing safety risks.
- Industry figures like Cohere CEO Aidan Gomez call the initiative an attempt to lock out competitors through barriers to entry and weaken open-weight AI.
- On the political side, President Donald Trump has dismissed AI takeover warnings as a hoax, while Democrats like Barack Obama and Kamala Harris support a slowdown and more transparency.
OpenAI, Anthropic, and Google want to slow down AI development, supposedly for safety reasons. Not everyone buys it.
Anthropic CEO Dario Amodei recently called on the industry and governments to coordinate a slowdown of frontier AI development, asking for antitrust exemptions to make it happen. Other major AI figures like Sam Altman and Elon Musk agreed. Critics are pushing back hard.
Cohere CEO calls it "a cartel by another name"
The most detailed rebuttal comes from Aidan Gomez, co-founder and CEO of AI company Cohere. In a blog post, he calls the proposal an attempt to set the rules and lock down the market. Gomez supports independent testing of powerful AI systems. But Anthropic's proposal goes much further: a handful of dominant labs would get an antitrust exemption, set shared standards, and force every other developer worldwide to follow their rules.
He's especially critical of the barriers to entry the proposal would create: massive compute, permanent monitoring infrastructure, dedicated safety organizations, embedded evaluation teams, and government connections deep enough to manage all of it.
The few evaluation organizations labeled "independent" are funded by the same companies they're supposed to audit. Anthropic's framework defines risk as a function of scale, Gomez argues. That makes the companies with the biggest systems the only ones deemed qualified to judge risk. "A wolf in sheep's clothing, a cartel by any other name," Gomez writes.
He sees the safety warnings as a smokescreen. While the industry debates hypothetical superintelligence, "voice cloning tools cheaper than a phone bill can empty a pensioner's account in minutes."
Hugging Face engineer calls it "bizarre nonsense"
Niels Rogge, a machine learning engineer at Hugging Face, a company whose business model depends on a broad selection of open AI models, calls the statements from Amodei and others "the most bizarre nonsense I've read lately." Amodei is mainly worried about Chinese models like those from DeepSeek and Z.ai and wants to restrict open-weight models to protect his competitive edge, Rogge argues. The AI labs are massive bets on scale, and smaller open-weight models could undermine that, though there aren't yet clear signs of this happening at any real scale.
Current LLMs are nowhere close to posing existential risks, Rogge says. He also points to the incident where OpenAI's autonomous models hacked Hugging Face, a closed-source player with enormous compute resources. Open-source models helped with the defense. That's exactly why the field needs transparency and openness, not closed doors, Rogge says.
Sacks tells AI labs to stop pretending they need permission
David Sacks, the White House AI and crypto czar under President Trump and a venture capitalist, responded on X directly to Amodei and Altman. By "any reasonable metric," OpenAI and Anthropic hold a duopoly in frontier intelligence, Sacks wrote. If their unreleased models are scary enough, he supports a voluntary slowdown.
The motivation isn't purely altruistic, though, Sacks argued. Both companies face massive product liability if their models enable a serious cyberattack. "The market already punishes models that behave in unpredictable or unauthorized ways. After the Hugging Face episode, it is simply good business for OpenAI and Anthropic to trade some raw power for reliability and predictability," Sacks wrote.
He also warned that tying the slowdown to a specific regulatory framework looks like extortion. China is "very unlikely" to join any global agreement, he added, and he might be right: Beijing has already rejected the proposed slowdown plans, dismissing them as a ploy for the US to cement its lead in AI.
Trump frames AI fears as a hoax
President Donald Trump went even further. In a post on Truth Social, he called warnings that "AI will take over the world" a "hoax," lumping them in with the climate change debate, the Russia investigation, and his impeachment proceedings. Trump also claimed that Chinese President Xi "just announced that China will be doing absolutely nothing to stand in the way of AI, or its future."
In several posts that followed, Trump doubled down. He called the "AI scam" pushed "by the Radical Left Dumocrats" reminiscent of "their Global Warming Scam of not so long ago, where everyone was going to die from extreme heat."
The CTO of the Department of War, Emil Michael, backed Trump up, saying the warnings come from "a deep-rooted ideology that is pretty dystopian." Michael has a deep, ideologically driven hostility toward Anthropic based on the company's ethical guidelines.
Political opponents of the Trump administration took the other side. Barack Obama, Bernie Sanders, and Kamala Harris all support a slowdown and have welcomed the labs' push for more transparency.
Cohere
Hugging Face / LinkedIn
David Sacks / X
Trump / Truth Social
Trump / Truth Social
Emil Michael / X
Obama / X
Sanders / X
Harris / X