Trump stakes US AI policy on a one-year lead over China

Trump said the US was ahead of China in AI by "a year, which is, you know, considered a lot" ^[4]. Evan Hubinger, Anthropic's alignment science lead, put the odds of AI killing all humans at greater than 10% over the next decade ^[8]. The two numbers answer different questions. A year is a tenth of a decade ^[18], so the asset the White House says it is defending covers a tenth of the horizon the risk estimate was made over.

A president's stated indifference sets enforcement appetite, appointments and tone. Statutes are Congress's work, and this record holds three bills ^[17]: Ted Cruz working with Senate colleagues on legislation to address "catastrophic risks" ^[15], the Ban Artificial Superintelligence Act from Bernie Sanders and Greg Casar, which would halt advanced AI development pending the creation of federal safety standards ^[11], and the FRONTIER Act from Jay Obernolte and Lori Trahan, intended to set rules for how advanced models can be deployed ^[12]. Quartz reports the warnings drew reactions from lawmakers in both parties ^[13].

Cruz's framing is the part worth underwriting. He told Forbes that "if there are gonna be killer robots, I'd rather they be American killer robots than Chinese killer robots" ^[16]. In that sentence the China argument and a catastrophic-risk bill are the same policy. Ron DeSantis posted on X that he found it alarming that "there is even a possibility that these technologies would elude human control" ^[14].

Jacob Coxon quit Anthropic earlier this week and posted publicly that both Anthropic and OpenAI were "racing straight to self-improving superintelligence and gambling with our lives" ^[5]. OpenAI chief scientist Jakub Pachocki said he was "concerned no one is prepared for the consequences of a continued rapid rise in machine intelligence" ^[10]. Those two hold posts at the companies whose spending a permissive federal posture protects, and their sentences are already public.

The permissive read holds if none of the three vehicles reaches a markup. Trump mentioned Xi Jinping's upcoming visit to Washington and said he expected the meeting to be "great," making no reference to AI talks, according to Forbes ^[7]. That is consistent with an administration that has nothing it plans to trade away on model development. The likelier version, or rather the more interesting one, is a Cruz-shaped compromise in which reporting and deployment duties are written into a bill that passes on the China frame ^[15][16]. A third path runs through the states, and the material here says nothing about preemption either way.

Of the three inputs an investor has, the executive posture is the least binding. Trump's answer on the tarmac was one sentence, "No, I don't have any" ^[2], and a deployment statute gets revised at considerably greater length. The thin part of this read is the bill text: none is quoted here, and Cruz's measure exists in the record only as work in progress with Senate colleagues ^[15]. If the FRONTIER Act dies in committee with no successor and no state rule replaces it, the permissive case wins.