US Treasury Secretary Scott Bessent announced that President Donald Trump and Chinese President Xi Jinping agreed that the final decision on the use of military artificial intelligence should always rest with a human being. This understanding is the highest-level signal yet of the two countries' intent to limit AI's decision-making authority in autonomous weapons systems. However, the ongoing chip export restrictions and the military AI investment race taking place in the same period cast a shadow over just how binding this understanding will be in practice.

What happened?

According to Bessent, Trump and Xi agreed on the principle that military AI systems cannot make lethal or strategic-level decisions without human approval. The announcement does not include a formal agreement text, signature, or oversight mechanism - it is based on the two leaders' verbal understanding. Which systems will fall under this principle, what will happen in case of violation, and whether the understanding will be put in writing all remain unclear.

Why does it matter?

The debate over autonomous weapons systems has been ongoing at the United Nations for years; due to objections from the US, Russia, and China, a binding international treaty still has not emerged. This is why the Trump-Xi understanding can be read as a bilateral, informal step forward in an area where multilateral negotiation has stalled - but being informal also means being unverifiable.

This is exactly where the contradiction emerges: the same two countries are waging a fierce export and technology war over AI chips. The US is restricting the sale of advanced AI chips to China; China, meanwhile, has made domestic chip production a military and strategic priority. While both sides are investing heavily in military AI infrastructure, their simultaneous acceptance of the 'human control' principle appears inconsistent. This may be less about reining in the arms race and more a diplomatic move serving to signal to the public and to each other that 'we are acting responsibly.'

Who does it affect, and how?

This understanding at the level of principle does not translate into a concrete short-term restriction for the defense industry. US defense technology companies and China's military AI suppliers will continue developing autonomous targeting and decision-support systems; if anything changes, it will be that these systems are marketed with the language of being 'human-approved.'

The real difference shows up in the lack of oversight and verification. The kind of independent inspection mechanisms that apply to nuclear weapons do not exist here; no one can verify the gap between a country saying 'there is human decision-making' and actually implementing it.

DimensionOfficial rhetoricActual situation
Decision mechanismHuman approval mandatoryNo oversight mechanism
Investment directionEmphasis on restrictionInvestment in military AI continues
Chip policySeparate agendaExport restrictions continue

The table shows that the same two governments are highlighting restrictive rhetoric on one hand while fueling competition on the other.

What's next?

The first question to watch is whether this verbal understanding will turn into a written framework. The second is whether it will become clear which systems the principle covers - areas like intelligence analysis, targeting, and cyber defense carry very different risk levels from one another. The third is whether chip export talks will proceed together with this military AI rhetoric or along a separate negotiating track.

For now, all that exists is a statement of intent. Until an oversight mechanism and a written commitment arrive, this understanding remains a diplomatic gesture; there is no concrete indication that it will slow either country's military AI investments or the chip war.