The upcoming meeting between U.S. Treasury Secretary Scott Bessent and Chinese Vice Premier He Lifeng arrives at a precarious moment. While diplomatic discussions on artificial intelligence are scheduled, President Donald Trump has publicly dismissed concerns about uncontrolled AI development as a “hoax.” Conversely, the state-owned China Daily has accused leading American AI firms of leveraging safety warnings to hinder China’s technological progress.
Neither nation appears willing to decelerate its AI ambitions, and mutual trust remains absent. However, experts argue that rapid technological advancements cannot wait for diplomatic frameworks to mature. The challenge lies in avoiding two common diplomatic failures: setting objectives so low that discussions become superficial, or pursuing such sweeping restrictions that agreement becomes unattainable due to lack of political trust and verification capacity.
A more viable near-term strategy involves parallel efforts in AI safety, anchored by technical exchanges that allow each country to improve system security according to its own standards. Additionally, establishing a protocol for communicating emerging AI incidents could prevent localized technical failures from spiraling into broader geopolitical crises.
Proposals include detailed technical dialogues on testing models for dangerous capabilities, such as those that could facilitate bioweapon development or evade human oversight. Both American and Chinese entities have identified these risks, with relevant safety testing protocols already appearing in Chinese technical standards. Sharing best practices in this area could enhance safety without compromising national security or competitive advantages.
This approach relies on self-interest rather than bilateral trust. As AI systems grow more powerful, both nations benefit from ensuring these technologies remain controllable. Unlike arms-control treaties, this diplomacy focuses on establishing shared operational procedures for a technology neither government fully understands.
Crucially, any technical dialogue must be accompanied by a government-to-government crisis-communication channel. Scenarios such as an American AI model hacking Chinese networks, or a Chinese drone attacking a U.S. vessel, require immediate clarification to distinguish between accidental malfunctions and deliberate acts. Defining notifiable incidents, designated contacts, and methods for signaling unauthorized actions without revealing sensitive intelligence will require rigorous technical diplomacy.
Experts recommend that initial incident reports be transmitted in writing rather than through real-time calls. Given China’s bureaucratic structure, written communication allows committees to analyze and formulate responses more effectively, reducing the risk of misinterpretation that can arise from oral exchanges. This method also mitigates the potential for recipients to distort messages.
While China has a history of suspending crisis-communication channels during periods of high tension, establishing a written protocol would at least remove the barrier of refusal to engage verbally. The talks should proceed without preconditions, keeping AI safety separate from broader issues like U.S. export controls on advanced chips.
The objective is not to transform intense competitors into partners, but to prevent the competition from yielding undesirable outcomes. With both nations far from the political conditions needed for broad agreements—Trump rejecting development slowdowns and Beijing viewing U.S. pacing as containment—a limited start is prudent.
Such modest beginnings could build momentum, creating expertise and interaction habits that become increasingly valuable if AI trajectories heighten concerns about severe risks. Eventually, both countries may find common ground in the understanding that while they need not agree on who wins the AI race, neither can afford to lose control of the technology.
Great analysis on keeping AI separate from chip bans. Hopefully, both sides remember that constraint before the next diplomatic row erupts.
Does anyone else find it alarming that we’re prioritizing crisis communication over actual alignment? We should stop building these things until we understand them.
Self-interest is the only glue holding this together. Without trust, we need cold, hard protocols, not hopeful diplomacy.
If Trump thinks AI safety is a hoax, how will the US even engage in these parallel technical exchanges? The timing seems impossible.
Writing things down is actually brilliant. It forces deliberation and prevents emotional, rushed reactions during a crisis.