总统唐纳德·特朗普在预定签署一项行政命令前数小时突然取消了该活动,该命令本将赋予政府在尖端AI模型公开发布前对其进行测试的权力。
据《纽约时报》解释,特朗普原本希望领先AI公司的高管能出席签署仪式。在得知部分CEO无法到场后,他决定取消活动。这让特朗普感到不快,尽管他只提前24小时通知了对方。其他迅速调整行程赶来的AI高管"在飞往白宫椭圆形办公室的途中"才得知此行已无意义。
Semafor的报道指出,OpenAI"支持"此次签署。然而,xAI创始人埃隆·马斯克和Meta CEO马克·扎克伯格据称协助"破坏"了这项行政命令,据传曾敦促特朗普"取消该命令"。此外,Semafor报道称,特朗普的前AI顾问大卫·萨克斯——据The Information指出,其政府特别雇员身份已于3月到期——也加入了推迟签署的推动行列。
据路透社报道,科技行业对该命令进行了游说反对,担心安全测试可能延迟模型发布或要求做出改变,从而阻碍模型开发。路透社指出,马斯克否认曾干预特朗普取消签署活动,他在X上发文称"这是假消息",并声称自己"不知道那项行政命令里写了什么"。
特朗普自重新执政以来对AI监管采取不干预态度,但在Anthropic指出其最新模型Mythos存在网络安全风险后,其政府成员感到担忧,开始建议进行安全测试。他们的计划是让特朗普扩大自愿接受政府测试和审查前沿模型的企业范围。
知情人士及获悉该行政命令的AI公司向The Information透露,特朗普政府与科技行业之间的一个分歧在于政府测试的时间表。政府希望在模型发布前最多90天进行评估,而AI实验室则主张将时间大幅缩短至仅14天。
据《纽约时报》报道,该行政令的目标是“让政府识别出人工智能模型所暴露的任何安全漏洞,并修补其系统中的问题,以帮助保护银行、公用事业和其他敏感行业免受网络攻击。”
特朗普正式向记者表示,他决定不签署该行政令,是因为他“不喜欢其中的某些方面”。他没有提供更多细节,但强调政府的安全测试可能会让美国在与中国的AI竞赛中落后,并声称“我真的认为(该行政令)可能成为一个阻碍因素。”
特朗普说:“我认为它会妨碍——你知道,我们正领先中国,领先所有人,我不想做任何会妨碍这种领先地位的事情。”
美国在AI安全竞赛中落后
目前尚不清楚特朗普是否计划重新安排该活动,也不清楚需要做出哪些改变才能确保他签署该行政令。
亚洲协会政策研究所中国分析中心研究员Lizzi C. Lee对《南华早报》表示,特朗普似乎正面临与中国相同的AI安全困境:如何在不妨碍前沿模型发展的前提下防范国家安全风险。
Lee认为,特朗普行政令的影响取决于“审查过程变得有多繁重”。如果特朗普的安全测试严格局限于国家安全领域,“那么它可能不会对领先的美国实验室造成太大影响,”Lee说。
据Lee称,与AI竞赛并行的还有“另一场可能更重要的竞赛”,即弄清楚“谁能在不扼杀创新的前提下治理强大的AI”。
在这场竞赛中,中国可能略微领先于美国。
《南华早报》的报道称,尽管美国在监管AI方面一直犹豫不决,但近几个月来,“中国的监管进程正在显著加速”。今年4月,北京发布了一项新规定,要求国内AI公司建立内部“人工智能伦理审查委员会”。今年5月,中国国务院(即内阁)概述了2026年的立法工作计划,旨在“完善AI治理,加快推动AI健康发展的全面立法”。
据《信息报》报道,在美国,不仅政党之间存在分歧,特朗普团队内部也并非铁板一块。据报道,紧张局势始于萨克斯的AI顾问任期突然结束后,这在美国白宫的AI领导架构中造成了“权力真空”。不过,《信息报》称,萨克斯仍会每周造访白宫。
据报道,特朗普正面临来自商务部及科技政策办公室的压力,要求其对AI监管保持轻触式态度,而像国家网络总监办公室这样更侧重安全的机构则认为,现在是时候进行治理了。
特朗普在美国AI安全方面的下一步行动,不仅会受到其政府内部担忧国家安全风险的官员的密切关注——这显然包括副总统JD·万斯,他在周三表示,在人们对Mythos提出担忧后,政府将“保护人们的数据”和“人们的隐私”列为优先事项。显然,中国也将期待特朗普定期提供AI安全方面的最新情况。
中国外交部证实,在最近与中国国家主席习近平的会晤中,特朗普同意“启动政府间AI对话”,以便在AI技术不断进步的情况下,共同应对新出现的国家安全风险。
President Donald Trump abruptly canceled an event on Thursday just hours before he was scheduled to sign an executive order granting the government the power to test frontier AI models before their public release.
As The New York Times explained, Trump had been hoping that top executives from leading AI firms would attend the signing. He decided to pull the plug after learning that some CEOs couldn’t make the event. That made Trump unhappy, even though he’d only given them 24 hours’ notice. Other AI executives who quickly rearranged their schedules to go “were midair on their way to the Oval Office” when they found out that the trip was for nothing.
Reporting from Semafor indicated that OpenAI “supported” the signing. However, xAI founder Elon Musk and Meta CEO Mark Zuckerberg reportedly helped “derail” the executive order, supposedly urging Trump to “call it off.” Additionally, Trump’s former AI advisor David Sacks—whose special government employee designation expired in March, The Information noted—joined the push to delay the signing, Semafor reported.
According to Reuters, the tech industry lobbied against the order, fearing that safety testing could delay model launches or require changes that set back model development. Musk has denied having a hand in Trump’s cancellation of the signing event, Reuters noted, writing on X that “this is false” and claiming that he doesn’t “know what was in that EO.”
Trump has taken a hands-off approach to regulating AI since retaking office, but members of his administration got spooked and began recommending safety testing after Anthropic flagged cybersecurity risks with its latest model, Mythos. Their plan was to get Trump to expand the number of firms submitting to voluntary government testing and vetting of frontier models.
Inside sources and AI firms briefed on the executive order told The Information that one tension between the Trump administration and the tech industry is the timeline for government testing. The government sought to evaluate models up to 90 days prior to release, while AI labs pushed for a much shorter timeline of only 14 days.
The EO’s goal, the NYT reported, was “for the government to identify any security vulnerabilities revealed by AI models and to patch problems in its systems to help protect banks, utilities, and other sensitive industries from cyberattacks.”
Officially, Trump told reporters that he decided against signing the order because he “didn’t like certain aspects of it.” He offered no further details but stressed that government safety testing could set the US back in the AI race with China, claiming that “I really thought [the order] could have been a blocker.”
“I think it gets in the way of—you know, we’re leading China, we’re leading everybody, and I don’t want to do anything that’s going to get in the way of that lead,” Trump said.
US lags behind in AI safety race
It’s unclear whether Trump plans to reschedule the event or what changes might be required to ensure he signs it.
Lizzi C. Lee, a fellow at the Asia Society Policy Institute’s Center for China Analysis, told the South China Morning Post that Trump appears to be navigating the same AI safety dilemma as China: how to guard against national security risks without slowing the development of frontier models.
Lee suggested that the impact of Trump’s order depends on how “heavy the review process becomes.” If Trump’s safety testing focuses narrowly on national security, “it probably won’t slow leading US labs much,” Lee said.
According to Lee, parallel to the AI race is “a separate, potentially more important race” to figure out how “who can govern powerful AI without choking off innovation.”
China may be slightly edging ahead of the US in that race.
SCMP’s report claimed that while the US has been hesitant to regulate AI, “China’s regulatory process is accelerating significantly” in recent months. In April, Beijing issued a new regulation requiring domestic AI firms to establish internal “artificial intelligence ethics review committees.” In May, the State Council, China’s cabinet, outlined a legislative work plan for 2026 to “improve AI governance and accelerate comprehensive legislation for the sound development of AI.”
In the US, discord exists not just between political parties but among Trump’s team, The Information reported. The tensions reportedly started after the abrupt end of Sacks’ tenure as AI advisor, creating a “power vacuum” within the White House’s AI leadership structure. Still, Sacks continues to visit the White House weekly, The Information reported.
As Trump reportedly faces pressure from the Commerce Department and the Office of Science and Technology Policy to maintain a light-touch approach to AI regulation, more security-focused agencies, like the Office of the National Cyber Director, think the time for governance is now.
Trump’s next moves on AI safety in the US won’t just be closely watched by officials within his administration concerned about national security risks—which apparently includes Vice President JD Vance, who said on Wednesday that the administration was prioritizing “protecting people’s data” and “people’s privacy” after concerns about Mythos were raised. Apparently, China will also be expecting regular AI safety updates from Trump.
At the recent summit with China’s president, Xi Jinping, Trump agreed “to launch an intergovernmental dialogue on AI” to mutually navigate emerging national security risks as AI technology advances, the Chinese Foreign Ministry confirmed.