WASHINGTON DC. WA, September 23, 2026 — Amid the always-bubbling U.S.-China tech rivalry, Washington and Beijing are preparing for a rare moment of cooperation. It will be their first dedicated AI safety dialogue between the two governments since President Donald Trump took office. The range of the talks is narrow, but their significance may be momentous, according to officials on both sides.
How the Talks Came Together
The dialogue was first set up during a May 2026 summit between Trump and Chinese President Xi Jinping. The October meeting follows what had been a nearly 2-year lull since the last of several formal U.S.-China discussions on AI, which took place in Geneva in 2024. They added that U.S. Treasury Secretary Scott Bessent would head up the American delegation.
Planned for days before Xi’s scheduled visit to Washington on Sept. 24, analysts say that Beijing frequently regards technical working sessions as preparation for high-level diplomacy. That timing hints that both governments hope to demonstrate tangible delivery before the two heads of state meet face-to-face.
What’s on the Table
The United States needs to focus discussions on observing AI-directed cyberattacks, officials briefed on the planning say. One idea raised by American negotiators is for AI labs in both countries to “police themselves” and report safety incidents. Officially, there has been no deal reached by either side, and expectations for more tangible results are low.
Concern likewise centers around shared issues about factor AI systems that can automate cyberattacks or discover weaknesses in the framework. Officials in both nations have indicated that increasingly sophisticated models could be abused by criminal organizations or adversarial actors. It is that common vulnerability, analysts say, which makes even small cooperation worth pursuing.
Rivalry That Hasn’t Gone Away
Although talks are taking place on the one hand, competition between the US and China regarding chips, semiconductors, and AI is still very pronounced. Export controls on high-end computing hardware are an ongoing dimension of the derailing relationship between Washington and Beijing. Neither government has indicated any openness to compromise on those other technology disputes.
The dynamic is “beyond geopolitical rivalry,” said Samm Sacks, a senior fellow at policy group New America. In both cases, she asserted that either country would be equally vulnerable to the dangers posed by mismanaged general AI systems, no matter which country built them first. Said she, “We are at a tipping point really, where frontier agents could do a lot of damage.”
Why Experts See an Opening
In a recent study, researchers at the Brookings Institution have argued that actually negotiating broad disputes is more aspirational than realistic, so better to settle for narrow, technical cooperation. They recommend focusing on common interests like avoiding AI-enhanced cyberattacks and improving emergency communication links. Getting working-level scientists and engineers involved, they say, might build trust that merely involving political leaders will not.
The mid-September dialogue may not yield long-term results as the U.S.-China AI cooperation should be appreciated more in history. Of course, even a small data-sharing agreement would represent a significant improvement over nearly years of very little on this front. For their part, both governments seem ready to explore whether safety concerns can happily exist next to a continuous strategic rivalry.
According to industry observers, the leading AI developers in both countries also recently made individual calls for greater and clearer international safety regulation. Others contend that a patchwork of regulations creates an environment in which models become increasingly powerful and invariably lend themselves to misuse. Even a narrow bilateral framework could provide the template for wider adoption by other nations at a point in time.
Skeptics warn that previous technology discussions between the two nations have resulted in more statements than enforceable commitments. They cite past cybersecurity deals that disappeared as non-cyber tensions climbed elsewhere in the relationship. That past softens hopes for what this round of negotiations can do.
However, officials involved in the planning say the stakes feel higher this time with how rapidly AI capabilities are progressing. Both governments have recognized the risks these unmonitored frontier systems create, which neither government can fully manage on its own. That mutual acknowledgment may be the greatest achievement of the talks to date, more significant than any particular proposal.
