#AI #China #Trump #regulation
The AI Slowdown Collides With Washington and Beijing
 
President Donald Trump is resisting stronger limits. He dismissed predictions of human extinction and estimated that the US may be about one year ahead of China, arguing that “whoever wins AI wins.” His June executive order lets developers provide the government with access to certain frontier models for up to 30 days before release, but participation is voluntary. It rejects mandatory licensing, permits and preclearance.
 
The administration remains divided. Treasury Secretary Scott Bessent and National Cyber Director Sean Cairncross have supported stronger safeguards, including discussion of a model-safety body based on FINRA. Trump adviser David Sacks argues that regulation would protect the largest laboratories from competition and give China time to catch up. He has challenged companies to slow themselves rather than make government coordination a condition for doing so.
 
Congress has reached little more than agreement that risks exist. House Speaker Mike Johnson and Democratic leader Hakeem Jeffries support guardrails, but neither has offered detailed legislation or a timetable. A government “kill switch” for rogue models is being discussed, yet comprehensive legislation has no realistic path through the House before the midterm elections.
 
Beijing’s response revealed the central tension in Amodei’s proposal. He argued that tighter controls on advanced chips, remote access to overseas data centers and unauthorized model distillation could widen the US lead over the next 3–5 years. That advantage would, in his view, allow Western laboratories to slow without letting China move ahead. The state-backed Global Times described the proposal as an attempt to preserve Washington’s technological dominance rather than a neutral safety plan. Chinese researchers argued that commercial pressure from lower-cost, open-source Chinese models may also have influenced Anthropic’s position.
 
Beijing is not dismissing AI risk itself. State Security Minister Chen Yixin issued his ministry’s first public warning about AI, citing deepfakes, automated troll networks, cyberattacks and the extraction of sensitive data. China’s cyberspace regulator has separately warned about an “extreme risk” of losing control of advanced systems. Yet Chinese officials frame the immediate danger mainly as foreign interference, cyber vulnerability and technological dependence—not the human-extinction scenarios discussed by some US researchers.
 
US and Chinese officials have been preparing for what would be the first bilateral talks devoted exclusively to AI safety during Trump’s second term. The proposed agenda includes monitoring AI-directed cyberattacks, sharing information about incidents and asking laboratories in both countries to police their own models. Treasury Secretary Scott Bessent could lead the US delegation, with Vice Premier He Lifeng or senior official Ding Xuexiang representing China. Chinese officials reportedly view the talks as an important deliverable for the planned September 24 summit, although the White House has said that no mid-September meeting is formally scheduled and the agenda and participants have not been finalized.
 
The summit itself is not guaranteed. Beijing has reportedly warned that Xi could cancel if Washington approves a proposed $14 billion arms package for Taiwan, although Trump said he was not concerned. The leaders must also address Iran, rare-earth supplies, technology export controls and a US–China trade truce that expires on November 10. AI safety is therefore becoming one bargaining item inside a much larger strategic negotiation—and neither country has a strong incentive to slow first.
 
“Pacing” will matter only if a company delays a valuable training run, discloses an embarrassing incident or accepts an independent evaluator’s refusal to approve a model. Until then, it means adding safety checks inside a race that continues to consume more capital and computing power. Investors may see little immediate effect on AI capex, but another serious failure could turn voluntary restraint into compulsory delay.