AI 监管之争尚未落幕

本周初，AI 领域的知名人士似乎——至少暂时如此——站到了支持 AI 监管的一边。上周末，Anthropic CEO Dario Amodei 提出了一项放缓 AI 开发的三步计划，包括让第三方评估人员进驻实验室、在美国国内行业间开展协调，以及在可能需要政府协助的情况下达成国际协议。OpenAI CEO Sam Altman、Google DeepMind 联合创始人 Demis Hassabis，甚至 SpaceX CEO Elon Musk，都曾公开表示认同这三项措施中的部分内容。

Anthropic 和 OpenAI 此前已经通过相关表态和公告暗示，它们与其他实验室正在制定某种行业框架。展现出支持监管的姿态，既是一次不错的公关行动，也可能帮助这些公司应对由逐步增多且愈发令人担忧的黑客攻击事件所引发的后果。

目前，政府参与这一进程看起来仍难以启动。特朗普政府抨击了 AI 存在安全危机的说法，将其称为一场“骗局”。AI 行业自身也存在分歧。Meta CEO Mark Zuckerberg 很快公开反对限制企业自主权；而本周早些时候，Wall Street Journal 报道称，Zuckerberg、Musk 和 Nvidia CEO Jensen Huang 阻挠了建立由行业出资、独立监管 AI 行业机构的提案——该机构将类似于金融业的私营非营利监管机构金融业监管局（FINRA）。但行业中的一些力量并未放弃这一设想，其中包括 OpenAI。

“人们希望知道，AI 是在安全地被开发，”OpenAI 全球事务主管 Chris Lehane 对 The Verge 表示，“这首先取决于像我们这样的公司自行采取的措施，但政府同样扮演重要角色。AI 政策的窗口已经打开，两党对针对前沿 AI 的强制性国家安全标准的支持正在增长。国会应当将这股势头转化为一个既能强化安全、支持美国创新，又能让美国继续保持 AI 领先地位的框架。”

一名因当前投资项目而要求匿名的 AI 风险投资人也同意监管是必要的。

“每种技术都会出现这样的时刻，无论是电力、汽车还是飞机：在某个阶段，行业必须联合起来，找到一种方式去回答，‘我们如何共同接受安全究竟意味着什么？’汽车曾经不安全，飞机曾经也不安全，随后我们不得不制定让所有人接受的安全标准，而这些标准实际上使技术得到更广泛的普及。但技术发展的早期阶段总是有些狂野且无序，技术就是这样演进的。”

Anthropic 拒绝置评，但 Amodei 在其文章中写道，该公司“始终……倡导经过审慎考虑的 AI 监管，即使这会让我们被指责为炒作、‘末日论’或监管俘获”。

其他公司则保持沉默。Safe Superintelligence Inc.、SpaceX 和 Google 均未回应置评请求。Thinking Machines Lab 拒绝置评。Meta 则援引了 Zuckerberg 发布的一则X 帖文。他在帖文中与放缓发展和加强监管的呼声保持距离，写道：“每一家实验室都有责任，也有动力，以安全训练模型所需的速度推进，并有能力采取自身行动，确保这一点实现。”

Zuckerberg 的思路目前似乎也受到特朗普的认同。特朗普上周因将近期对 AI 风险的担忧称为“骗局”而登上新闻头条。他在 Huang 于一场会议舞台上发言时致电 Huang，并通过扬声器对现场人群表示：“机器人不会接管一切。”同一天，他还在社交媒体发帖称，AI 唯一需要的控制或“护栏”，就是一位“强大而聪明（高 IQ！）的总统，而美国恰好拥有这样一位，而且绰绰有余！”他补充称，特朗普政府已经利用现有的、针对这些公司的“强大刑事和监管权力”，阻止了 AI “人士”做坏事或潜在的坏“事”。他谴责一场针对 AI 和数据中心的“病态阴谋”，并称谁赢得 AI，谁就赢得全局。“阴谋论者、叛国者、内奸和泄密者，当心！”他写道。

对于纽约大学兼职教授、美国国土安全部前新兴技术政策主管 Nick Reese 而言，特朗普目前反对 AI 监管的言论令人困惑，因为 Reese 曾与第一届特朗普政府合作制定第 13960 号行政命令——特朗普签署了这项命令，且该命令尚未被废除。这项命令的核心，是为联邦政府使用 AI 制定一套原则，以保护“隐私、公民权利及公民自由”。

“六年前，特朗普政府签署了关于 AI 安全的行政行动，因此总统在这个问题上的立场确实发生了变化，”Reese 表示，“AI 安全需要的远不止一名民选官员的智力能力。”

正如 AI 行业博主 Zvi Mowshowitz 在一则X 帖文中所说，特朗普未必完全反对监管——毕竟，是他对前沿实验室施加了一些限制，例如要求进行发布前测试以及接受政府测试。但按 Mowshowitz 的说法，特朗普更多是受“感觉”而非一贯政策驱动。“特朗普还没有接受这种‘生存风险’说法，那是‘负面力量’，”Mowshowitz 写道。因此，政府护栏并非完全不可能出台，但其形式可能与 OpenAI 或 Anthropic 所希望的方案相去甚远，更不必说力量较弱、独立的 AI 安全倡导者所提出的计划。

Reese 表示，整体来看，政府在 AI 监管上的立场“一直摇摆且不一致”。“其结果是，我们如今将 AI 安全置于‘机器人要来抓我们’的末日情境中讨论。但把问题框定为‘机器人要来抓我们’，实际上严重扭曲了真正的核心问题。核心问题是，当前已经有并非实体化于机器人之中的 AI 正在伤害人类。”

特朗普在 Truth Social 的帖文中写道，政府目前对 AI 实验室同时拥有刑事和监管权力；但 Reese 表示，我们距离对 AI 行业拥有足够的监管授权还相差甚远。随着技术发展，以及公众持续反弹——包括针对数据中心的抗议、抵制服务的行动，以及对 AI 行业权力集中于少数人之手的抗议声浪——要求监管的呼声只会继续增加，这种压力可能影响特朗普目前的立场。

Reese 表示，对监管的反对可归结为一种“过时”的观点，即监管总会不利于创新。相反，他认为，重要的是思考任何事物为何会受到监管。

“我们监管那些绝不能容忍失败的事物……例如金融业或航空业，”他说，“我们建立系统、检查机制、质量保证和后备机制，以确保这些事物不会危害生命，也不会对我们造成系统性伤害。但不知为何，我们却让 AI 获得了豁免。”

At the start of this week, the who’s-who of AI seemed — at least tentatively — on the side of AI regulation. Over the weekend, Anthropic CEO Dario Amodei had proposed a three-step plan for slowing AI development, including by embedding third-party evaluators in labs, coordinating across the domestic industry, and forging international agreements potentially with government assistance. OpenAI CEO Sam Altman, Google DeepMind co-founder Demis Hassabis, and even SpaceX CEO Elon Musk publicly seemed to agree on aspects of all three things.

Anthropic and OpenAI had already been dropping hints that they and other labs were working on some kind of industry framework. Appearing pro-regulation looked like both a good PR move and, potentially, a way for companies to address fallout from progressively more concerning hacking incidents.

For now, any government involvement in this process looks like a non-starter. The Trump administration has attacked the idea of a safety crisis in AI, calling it a “hoax.” The AI world itself is also divided. Meta CEO Mark Zuckerberg quickly came out against limiting companies’ autonomy, and earlier this week, the Wall Street Journal reported that Zuckerberg, Musk, and Nvidia CEO Jensen Huang scuttled proposals for an industry-funded independent regulator for the AI industry — akin to the ​Financial Industry Regulatory Authority (FINRA), a private nonprofit regulator in finance. But parts of the industry aren’t giving up on the idea, including OpenAI.

“People want to know AI is being developed safely,” Chris Lehane, OpenAI’s global affairs chief, told The Verge. “That begins with the steps companies like ours take on our own, but government has an important role too. The AI policy window is open, with growing bipartisan support for mandatory national safety standards for frontier AI. Congress should turn that momentum into a framework that strengthens safety, supports American innovation, and keeps the U.S. leading in AI.”

An AI venture capitalist, who requested anonymity due to their current investments, agreed that regulation was necessary. “You see these moments in every technology, whether it was electricity, cars, or airplanes, where, at some point, the industry had to come together to find a way to say, ‘How do we collectively embrace what safety means?’ Cars were unsafe, airplanes were unsafe, and then we had to create safety standards that everybody embraced that actually made technology more widespread. But the early days are always a little wild and unruly, and that’s how technology unfolds.”

Anthropic declined to comment, but Amodei wrote in his essay that the company has “always … advocat[ed] for well-considered regulation of AI, even when this gets us accused of hype, ‘doomerism’, or regulatory capture.”

Other companies have remained quiet. Safe Superintelligence Inc., SpaceX, and Google didn’t respond to a request for comment. Thinking Machines Lab declined to comment. And Meta pointed to an X post by Zuckerberg, in which he distanced himself from the calls for a slowdown and regulation, writing, “Every lab has the responsibility and incentive to move at the pace required to train its models safely, and the ability to take its own actions to ensure that happens.”

Zuckerberg’s line of thinking appears in vogue with Trump right now. The president made headlines last week for calling recent fears about AI risks a “hoax.” He called Huang while the latter was onstage at a conference and told a crowd over speakerphone that “The robots will not be taking over.” That same day, he also posted on social media that “the only control or ‘guardrails’ that AI needs is a STRONG AND SMART (High IQ!) PRESIDENT, and the U.S.A. has that, in spades!” He added that the Trump administration has already “stopped AI ‘people’ from doing bad, or potentially bad, ‘things,’” using its existing “tremendous CRIMINAL and REGULATORY power over these companies!” He denounced a “SICK conspiracy” going on against both AI and data centers and that whoever wins AI wins overall. “Conspiracy Theorists, Treasonists, Traitors, and Leakers, BEWARE!” he wrote.

For Nick Reese, an adjunct professor at New York University and the Department of Homeland Security’s former director of emerging tech policy, Trump’s current anti-regulation statements about AI are confusing because he worked with the first Trump administration to develop Executive Order 13960 — which Trump signed and has not been repealed. The order was centered on creating a set of principles for federal AI use that would protect “privacy, civil rights, [and] civil liberties.”

“Six years ago, the Trump Administration signed executive action on AI safety, so there is a real change in the president’s position on the topic,” Reese said. “AI safety requires more than the intellectual abilities of one elected official.”

As AI-industry blogger Zvi Mowshowitz put it, Trump isn’t necessarily fully anti-regulation — after all, he’s the one who slapped some restrictions on frontier labs, like pre-release testing by the government. But he’s driven more by, in Mowshowitz’s words, “vibes” than consistent policy. “Trump’s not yet buying this ‘existential risk’ thing, that’s ‘negative forces,’” Mowshowitz wrote. So government guardrails aren’t exactly ruled out, but they may bear little resemblance to what OpenAI or Anthropic want — let alone the plans of less-powerful, independent AI safety advocates.

Overall, the administration’s positions on AI regulation have been “wobbly and inconsistent,” Reese said. “The result of that is that we are now framing AI safety in this doomsday scenario where the robots are coming to get us. But framing it as ‘robots are coming to get us’ really distorts what the core problem actually is. The core problem is that we have AI that is not physical or embodied in a robot that is hurting humans now.”

Trump wrote in his post on Truth Social that the administration currently has both criminal and regulatory power over AI labs, but Reese said we’re nowhere near sufficient regulatory authority over the AI industry. As the technology advances, and the general public continues to push back — protests, boycotts, and outcry over the AI industry’s concentrated power in the hands of few — calls for regulation are only increasing, and that pressure could influence Trump’s current position.

Reese said the opposition to regulation boils down to a “tired” idea that it’s always bad for innovation. Instead, he says, it’s important to think about why anything gets regulated.

“We regulate things that we can tolerate no failure on … like the financial sector or airlines,” he said. “We build systems and checks and quality assurance and fallback mechanisms to make sure that those things don’t harm life or harm us systemically. But somehow we’re exempting AI from that.”

Follow topics and authors from this story to see more like this in your personalized homepage feed and to receive email updates.

• Hayden Field

• AI

• Anthropic

• Elon Musk

• Google

• Law

• OpenAI

• Policy

• Report

• Tech

• xAI