San Francisco’s Dreamforce conference is bringing Nvidia, OpenAI, and Anthropic together this week as Silicon Valley argues over whether frontier AI is moving too fast. President Trump made his position clear Monday. During a call with Nvidia CEO Jensen Huang at the All-In Summit in Los Angeles, Trump dismissed the warnings as a “hoax.” Huang agreed.

But the more consequential AI conversation comes next week in Washington. Chinese President Xi Jinping is scheduled to meet Trump on September 24 for their second summit this year. The question facing them is not whether the AI competition can be stopped. It is whether the world’s two leading AI powers can compete without allowing that competition to become dangerously uncontrolled. U.S. officials are already exploring talks with Beijing on AI-directed cyberattacks, information sharing, and China’s alleged distillation of American models.

In “The New AI Cold War,” I argued that artificial intelligence would become strategic terrain in the U.S.-China rivalry. The contest would involve far more than who builds the smartest model. It would turn on chips, compute, energy, standards, military applications, and digital infrastructure, with the older struggle between liberty and authoritarian control running beneath it. That competition is now shaping national policy in both countries.

Neither Washington nor Beijing can safely assume the other will slow first. Any restraint that binds one side but not the other risks becoming a strategic concession. Trump therefore faces a harder challenge than choosing between a pause and an unrestricted race: America must remain ahead while establishing a few rules that reduce dangers neither country can afford to ignore.

Beijing is building capability across chips, frontier models, industrial applications, military systems, and global technical standards. As I warned in “The New AI Cold War,” China does not need the world’s single best frontier model to gain global strategic influence. Lower-cost, capable Chinese open-weight models give Beijing another path to influence by encouraging governments, companies, and developers abroad to build on Chinese technology. An American pause would not create a global pause; it would simply hand Beijing strategic space.

Competition, however, cannot become an excuse for operating without safeguards. Recent AI cyber incidents and concerns about capable autonomous systems demonstrate why Washington and Beijing share at least some interests even while competing fiercely.

The September 24 summit should therefore focus on a few practical rules that manage the competition without pretending it can be suspended.

Trump and Xi should first establish a permanent emergency channel between the two governments for serious AI incidents. During the first Cold War, Washington and Moscow learned that communication between adversaries could stop an accident from becoming a catastrophe. A permanent U.S.-China AI incident channel could provide the same protection when a major cyber event, dangerous model behavior, or uncertainty about another country’s actions threatens escalation.

They should also establish rapid notification procedures for serious cross-border AI incidents, including major cyber intrusions or an unexpected loss of control over an advanced system. Neither country needs to surrender proprietary models, source code, or intelligence. But if a system escapes containment, enables a major cyberattack, or reveals a capability that could destabilize both nations, silence serves no one.

Third, they should direct technical teams to build common testing standards for the narrowest, highest-consequence capabilities: AI-enabled cyber operations, biological risks, and the use of AI in nuclear warning and command systems. Cooperation need not mean trust. Arms-control history proves that rivals can build safeguards precisely because they don’t trust each other.

Finally, both presidents should affirm that artificial intelligence will never hold final authority to authorize or initiate the use of nuclear weapons. AI may support analysis, warning, and defense, but decisions carrying civilization-scale consequences must remain under responsible human authority.

The summit cannot substitute for American policy at home. Speaker Mike Johnson (R-La.) said Sunday that an AI moratorium would hand China the advantage, while calling ideas like a federal kill switch and mandatory pre-release safety review “great ideas” he has yet to bring to a vote. Congress should turn that instinct into action: establish narrow rules for serious frontier model incidents, protect American compute and intellectual property, and leave smaller companies and open-weight developers room to compete.

Industry has obligations as well. Frontier companies should share serious incident information, strengthen model and data center security, and accept credible independent testing of their most capable systems. At the same time, America must expand the foundations of technological leadership: energy, semiconductors, data centers, scientific research, skilled workers, and competitive American models. 

The competition reaches well beyond model capability. Infrastructure, military advantage, economic adoption, global influence, and control will all help determine who shapes the AI age.

Scripture joins strength with judgment: “for by wise guidance you can wage your war, and in abundance of counselors there is victory” (Proverbs 24:6, ESV). The AI competition demands that same combination of strength and prudence.

Trump should not ask Xi to end the AI Cold War. Xi will not agree, and America should not surrender its advantage. But both leaders have reason to prevent competition from producing failures neither nation can control.

America should lead the AI age. Leadership, however, requires more than moving fastest. It requires the strength to compete, the judgment to establish limits where catastrophe serves no one, and the confidence to keep human beings responsible for the machines they build.