内容来源：https://www.wired.com/story/trumps-ai-safety-accord-is-a-fancy-pinky-swear/

内容总结：

多家美国人工智能公司高管签署白宫人工智能安全协议，但约束力有限

本周二，谷歌、Anthropic、Meta、OpenAI、xAI和英伟达等六家美国人工智能企业的高管共同签署了《白宫超级智能协议》。该协议在总统特朗普主持的午餐会后对外公布。特朗普称此举是“极大的自我监管”，但该协议并非这些公司内部人士此前呼吁的全面人工智能监管方案。即便企业未能履行这些公开承诺可能构成对消费者的非法欺骗，也不足以触发大规模改革。

协议声明称，企业“应当落实”以下几项内容：第一，确保建立“强有力的内部控制”，以监测其模型的能力，防止模型失控或以意外方式入侵系统；第二， “赋权”内部团队开展上述工作，并在出现问题时进行补救；第三，鼓励企业与外部监督机构合作，由后者进行独立评估，以确认上述工作切实到位；第四，确保董事会设立委员会，接收有关上述工作的报告。

这些自愿性承诺在一定程度上表明，前沿实验室正在努力探索如何打造对公众安全的产品。然而，如果有任何人寄望于获得美国政府的全面背书，这一期望很快就被打破——有消息称，美国联邦贸易委员会（FTC）打算就潜在的消费者保护问题对多家AI公司展开调查。

这份声明是数周来围绕人工智能监管争论的最终产物。此前，前沿AI实验室一直在寻求反垄断豁免，以便在不受监管审查的情况下就安全工作进行协调。（值得注意的是，反垄断专家和从业者表示，这种豁免并无必要，因为AI终结人类——研究人员警告称若不采取行动这将不可避免——本身就可被视为反竞争行为，毕竟人类是任何市场不可或缺的组成部分。）

围绕这份声明的舆论铺垫已持续多时，但这并非这些公司首次与政府签署关于AI安全实践的自愿协议。2025年初，英国和韩国曾宣布一套“前沿AI安全承诺”，其中包括对AI模型就“严重且新型威胁”进行内部和外部红队测试的承诺，以及“致力于在AI安全问题上实现信息共享”。

谷歌、Anthropic、Meta、OpenAI、xAI和英伟达均未回应置评请求。白宫也未立即作出回应。

FTC前首席技术官尼尔·奇尔森在X平台上发文评论该协议称：“如果某家公司实质性违反了这些承诺中的任何一项，FTC有可能对此类保证加以执行。”

但那些仅凭该协议就期待达成全面和解、并对失控的“超级智能”进行大刀阔斧约束的人，可能会大失所望。

（披露：本文作者此前曾在FTC技术办公室工作，于2025年11月辞职。）

未能兑现公开承诺可能被视为欺骗性商业行为，违反《联邦贸易委员会法》。当FTC指控某公司欺骗消费者时，其通常采取的补救措施是要求该公司承诺不再撒谎。

例如，FTC近期宣布与三家公司达成和解。FTC指控这三家公司吹捧一项服务，声称该服务可帮助企业基于从智能手机、智能电视及其他设备收集的音频录音来投放定向广告。FTC表示，这些公司就自身窥探消费者的能力向企业客户撒了谎。解决方案是什么？这些公司不再被允许就其广告服务以及消费者语音数据的收集和使用作出虚假陈述。该案并未涉及从手机收集语音数据是否合法的问题，和解协议也没有为这三家公司未来真正实施此类行为设定任何护栏（除了不得撒谎之外）。

OpenAI对其智能体多次入侵他人系统的次数以及自身需要改进安全实践一事表现得相当坦诚。但正如专家向《连线》杂志指出的那样，美国政府即使最终决定不采取执法行动，也可以对企业展开调查。（例如，据报道FTC在2023年曾向OpenAI发出民事调查令，但后续进展不明。）

在周二与特朗普的AI午餐会后，《纽约邮报》报道称，FTC正计划对Anthropic、OpenAI、其他未具名前沿实验室以及一家名为METR的第三方评估非营利组织展开“大规模调查”。METR及各实验室均未回应置评请求。

《纽约邮报》称，FTC尚未向预定调查对象发出民事调查令——这通常是任何调查最早采取的步骤之一。FTC发言人证实确实存在某种调查，但拒绝提供更多细节，包括涉及哪些消费者保护问题。

FTC主席安德鲁·弗格森的名字出现在“超级智能”午餐会的座次表上。他此前曾解释称，执法者并非从一开始就掌握完整的案件。“你要不断寻找烟雾，看是否有火，”他去年对一群反垄断爱好者如是说。

即便FTC调查继续推进并确实发现了违规行为，也有人怀疑最终是否会真正对AI公司采取行动。

“FTC正在调查这些公司，这对担心AI的美国人来说应该是件令人鼓舞的事，”曾在莉娜·汗担任主席期间任FTC公共事务主管的道格拉斯·法拉尔表示。“但我们要记住，弗格森虽然是FTC主席，但真正说了算的是特朗普。在昨天他与AI CEO好友们的欢聚之后，我怀疑特朗普不会允许对这些公司采取任何有牙齿的法律行动。”

中文翻译：

周二，谷歌、Anthropic、Meta、OpenAI、xAI和英伟达的高管共同签署了《白宫超级智能协议》，该协议在唐纳德·特朗普总统举办的午餐会后公布。

用特朗普的话说，这是一次“极大的自我监管”行动，但该协议并非各公司内部人士一直呼吁的那种全面的人工智能监管。而且，即便未能履行这些公开承诺可能构成对消费者的非法欺骗，那也不足以触发大规模改革。

声明称，各公司“应当实施”以下几项措施：第一，应确保拥有“强有力的内部控制”，以监控其模型的能力，以及防止模型失控或以非预期方式入侵系统的能力。第二，应“赋权”一个内部团队来执行这项工作（并在出问题时进行补救）。第三，鼓励各公司与外部监督机构合作，并“赋权”该机构进行独立评估，以确保上述工作切实到位。最后，应确保其董事会设立一个委员会，负责接收关于所有这些工作的报告。

这些自愿性承诺在一定程度上表明，前沿实验室正在努力探索如何制造对公众安全的产品。然而，如果有人认为这能换来美国政府的全面背书，那么这种期望在消息传出后迅速破灭——联邦贸易委员会打算以潜在的消费者保护问题为由调查多家AI公司。

这份声明是数周来围绕AI监管辩论的最终产物。此前，前沿AI实验室一直在寻求反垄断豁免，以便在不受监管审查的情况下协调安全工作。（值得注意的是，反垄断专家和从业者表示，这种豁免并无必要，因为AI终结人类——研究人员警告称若不采取行动这将不可避免——可以说本身就是反竞争行为，毕竟人类是任何市场不可或缺的组成部分。）

围绕这份声明有诸多铺垫，但这并非这些公司首次与政府签署关于AI安全实践的自愿协议。2025年初，英国和韩国宣布了一套“前沿AI安全承诺”，其中包括对AI模型针对“严重和新颖威胁”进行内部和外部红队测试的承诺，以及“致力于在AI安全问题上实现信息共享”。

谷歌、Anthropic、Meta、OpenAI、xAI和英伟达均未回应置评请求。白宫也未立即回应。

联邦贸易委员会前首席技术官尼尔·奇尔森在X上发帖评论该协议称：“如果某家公司实质上未能兑现其中任何一项承诺，这类保证是联邦贸易委员会有可能强制执行的。”

但那些仅凭该协议就期待全面解决方案和戏剧性遏制失控“超级智能”的人，可能会大失所望。

（披露：本文作者此前曾在联邦贸易委员会技术办公室工作，于2025年11月离职。）

未能兑现公开承诺可能被视为欺骗性商业行为，违反《联邦贸易委员会法》。当联邦贸易委员会指控某公司欺骗消费者时，其通常给出的补救措施是要求该公司承诺不再撒谎。

例如，联邦贸易委员会最近宣布与三家公司达成和解，此前该机构指控这三家公司吹捧一项可帮助企业基于从智能手机、智能电视和其他设备收集的录音进行广告定向投放的服务。该机构称，这些公司就其窥探消费者的能力向企业客户撒了谎。解决方案是什么？这些公司不再被允许对其广告服务以及消费者语音数据的收集和使用作出虚假陈述。该案并未涉及从手机收集语音数据是否合法的问题，和解协议也没有为这三家公司中任何一家将来真的想这么做设定任何护栏（除了不得撒谎之外）。

OpenAI对其智能体侵入他人系统的次数以及为何需要改进安全实践一直相当坦率。但正如专家向《连线》杂志指出的那样，即使美国政府最终决定不采取执法行动，也可以对公司展开调查。（例如，据报道联邦贸易委员会曾在2023年向OpenAI发出民事调查令，但后续进展不明。）

在周二与特朗普的AI午餐会后，《纽约邮报》报道称，联邦贸易委员会正计划对Anthropic、OpenAI、其他未具名前沿实验室以及一家名为METR的第三方评估非营利组织展开“全面调查”。METR和各实验室均未回应置评请求。

《纽约邮报》称，该机构尚未向预定目标发出任何民事调查令——这通常是各类调查的第一步。联邦贸易委员会发言人证实确实存在某种调查，但拒绝提供任何额外细节，包括涉及哪些消费者保护问题。

联邦贸易委员会主席安德鲁·弗格森的名字出现在“超级智能”午餐会的座次表上，他此前曾解释称，执法者并非总能一开始就掌握完整的案情。“你要不断寻找烟雾，看是否有火，”他去年对一群反垄断爱好者说。

即使联邦贸易委员会的调查继续推进并确实发现了违规行为，也有人怀疑这最终能否对AI公司采取任何行动。

“联邦贸易委员会正在调查这些公司，这对担心AI的美国人来说应该是件令人鼓舞的事，”曾在莉娜·汗担任主席期间任联邦贸易委员会公共事务主管的道格拉斯·法勒说。“但我们要记住，弗格森或许是联邦贸易委员会的主席，但真正说了算的是唐纳德·特朗普，我怀疑在昨天与他那些AI CEO好友的欢乐聚会之后，特朗普不会允许对这些公司采取任何有牙齿的法律行动。”

评论

返回顶部

英文来源：

On Tuesday, executives for Google, Anthropic, Meta, OpenAI, xAI, and Nvidia all signed “The White House Accord on Super Intelligence,” which was announced following a luncheon held by President Donald Trump.
The effort was an act of “tremendous self-regulation,” as Trump put it, but the agreement is not the sweeping AI regulation that people inside the companies themselves have been asking for. And while failing to honor these public commitments could count as illegally lying to consumers, even that wouldn’t be enough to trigger massive reforms.
The statement says that companies “should implement” several things: First, they should ensure they have “robust internal controls” to monitor their models’ capabilities and ability to not go rogue or hack things in unintended ways. Second, they should “empower” an internal team to do that work (and remediate when things go wrong). Third, it encourages them to partner with an external monitor who is “empowered” to do independent assessments to make sure that all of that work is happening. And finally, they should make sure their board of directors has set up a committee to receive reports about all of these efforts.
The voluntary commitments are some indication that the frontier labs are making an effort to figure out how to make products that are safe for the general public. However, if there were any hopes for a blanket blessing from the US government, those were quickly dashed after news broke that the Federal Trade Commission intends to investigate several AI companies over potential consumer protection issues.
The statement is the culmination of weeks of debate over AI regulation. It also follows frontier AI labs seeking an antitrust exemption so they can coordinate on safety work without regulatory scrutiny. (Notably, antitrust experts and practitioners say such an exemption is unneeded because AI ending humanity—which researchers have warned will come about if nothing is done—is arguably anti-competitive because humans are an essential part of any given marketplace.)
There’s been a lot of build-up to this particular statement, but this is not the first time that these companies have signed a voluntary agreement about their AI safety practices with a government. In early 2025, the United Kingdom and the Republic of Korea announced a set of “Frontier AI Safety Commitments,” which included promises about doing internal and external red-teaming of AI models for “severe and novel threats” and also “to work toward information sharing” on AI safety issues.
Google, Anthropic, Meta, OpenAI, xAI, and Nvidia did not respond to requests for comment. The White House also did not immediately respond.
Neil Chilson, a former chief technologist for the FTC, posted on X of the accords that “this is the kind of pledge that the FTC could potentially enforce, if a company materially failed to follow through on any of these promises.”
But those hoping for a sweeping settlement and a dramatic reining-in of rogue “super intelligence” based on the accords alone might find themselves disappointed.
(Disclosure: This story’s author previously worked in the Office of Technology at the FTC and resigned in November 2025.)
Not following through on a public promise could be considered a deceptive business practice and a violation of the FTC Act. When the FTC alleges that a company deceived its customers, the remedy it usually prescribes is making the company promise not to lie again.
For example, the FTC recently announced it had settled with three companies after it alleged they were touting a service that could help businesses target ads based on audio recordings collected from smartphones, smart TVs, and other devices. The companies, the agency said, lied to businesses about their ability to creep on consumers. The fix? The companies are no longer allowed to misrepresent their advertising services and the collection and use of consumers’ voice data, among other things. The case does not address whether it’s OK to collect voice data from your phone, and the settlement doesn’t set out any guardrails if one of the three companies wanted to actually do that in the future (other than not lying about it.)
OpenAI has been remarkably open about the number of times its agents have broken into other people’s systems and how it needs to improve its safety practices. But, as experts have pointed out to WIRED, the US government can open investigations into companies even if they eventually decide there’s no enforcement action to bring. (The FTC, for example, reportedly sent a civil investigative demand to OpenAI in 2023, but it’s unclear what came of that.)
Following Tuesday’s AI luncheon with Trump, the New York Post reported that the FTC was planning a “sweeping probe” into Anthropic, OpenAI, other unnamed frontier labs, and a nonprofit called METR that does third-party evaluations. METR and the labs didn’t respond to a request for comment.
The Post says that the agency still hasn’t issued any civil investigative demands to its intended targets—usually one of the first steps in any sort of investigation. An FTC spokesperson confirmed that there is some sort of investigation, but they declined to provide any additional details, including what consumer protection issues were of concern.
FTC chairman Andrew Ferguson, who was included on the “super intelligence” luncheon’s seating chart, has previously explained that enforcers don’t always have a fully baked case from the get-go. “You're constantly looking for smoke to see if there's fire,” he told a crowd of antitrust enthusiasts last year.
Even if the FTC investigation moves forward and the agency does identify violations, some are skeptical that it will result in any action against the AI firms.
“It should be encouraging to Americans worried about AI that the FTC is investigating these companies,” says Douglas Farrar, who was the head of public affairs for the FTC during Lina Khan’s tenure as chair. “But let’s remember that Ferguson may chair the FTC, but Donald Trump runs it, and I doubt after yesterday's love fest with his AI CEO buddies that Trump will allow any legal action with teeth against these companies.”
Comments
Back to top