US and China open channel for AI incidents
The two biggest AI powers want to warn each other about incidents posing risks to national security.
The United States and China have agreed to establish a communications channel for serious AI incidents. The precise procedures, reporting threshold and powers have not yet been made public.
The channel follows talks between American and Chinese officials around the meeting of President Donald Trump and President Xi Jinping in Washington. After the summit, the White House said that both countries wanted to establish a bilateral channel for AI incidents.
US Treasury Secretary Scott Bessent had earlier said that Washington had proposed a reporting mechanism for events that could have implications for national security. This would cover not only errors in models, but potentially also incidents in which AI systems are involved in cyberattacks, military decision-making or other sensitive infrastructure.
The announcement does not yet amount to an AI safety treaty. Little is publicly known about which agencies will contact each other, how quickly reports must be made and whether companies will be required to report incidents. It is also unclear which Chinese regulators or government departments will take part.
The step comes as Washington and Beijing continue to treat each other as strategic and economic competitors. Both countries are trying to give their own companies an advantage in chips, data centres and generative AI. That competition makes information-sharing politically sensitive, especially when an incident may also contain indications of espionage or sabotage.
A direct channel could limit misunderstandings if an autonomous system displays unexpected behaviour or if a cyber incident is wrongly attributed. The channel does not thereby resolve the underlying differences over export restrictions, oversight and how much room companies should be given to launch new systems quickly.
One story, several perspectives
What is established
- The United States and China want a bilateral channel for serious AI incidents.
- The proposal focuses on incidents that could have national security implications.
- The countries remain strategic and economic competitors at the same time.
- There is still no publicly available detailed protocol setting out reporting thresholds and powers.
Left
Arguments A channel is useful, but insufficient as long as companies and governments continue to scale up their systems without binding safety standards. International rules, transparency and human control should take precedence over the race for market share.
Values Public safety, democratic oversight, international cooperation and limiting the concentration of private power.
Consequences Without binding rules, the channel could primarily serve as crisis communication after damage has already occurred. Stricter standards could reduce the risk of escalation and misuse, but innovation could proceed more slowly.
Centre
Arguments A limited reporting line is a realistic first step. The countries do not need to agree on every aspect of AI to make arrangements for incidents that could escalate quickly.
Values Caution, institutional reliability, proportionality and practical cooperation.
Consequences A small channel could reduce misunderstandings, but only if it later becomes clear who assesses reports and what follow-up steps are possible.
Right
Arguments The US must maintain its technological lead and prevent safety arrangements from becoming a back door for Chinese influence or American regulation. Contact is useful to prevent miscalculations, but national security and freedom to innovate must remain the guiding principles.
Values National sovereignty, competitiveness, security and limited government interference.
Consequences A narrow channel could keep risks manageable without bringing AI development to a halt. Overly broad arrangements could disadvantage American companies and expose strategic information.
The perspectives describe how these political currents typically approach the subject; the newsroom takes no position on which perspective is right.
Fact-check Approved · Nour Haddad — AI agent
This check was carried out by AI: every claim was re-tested against the sources. Even an approved article can contain errors — stay critical.
The reporting supports the existence of an American-Chinese reporting mechanism and the national security context. Unknown details have not been presented as established facts.
- confirmed The United States and China want to establish a communications channel for AI incidents. — The White House reported this after the summit, and AP described it. source
- confirmed Scott Bessent earlier proposed a reporting mechanism for incidents with national security implications. — AP quotes Bessent on the American proposal. source
- confirmed The precise procedures and powers have not been made public. — Public reporting describes the proposal and the intention, but not a detailed protocol. source
- confirmed Xi said that AI must remain under human control. — CBS News reported this statement during the summit. source