U.S. and China Prepare for High-Stakes Bilateral AI Safety Talks in Mid-September

The United States and China are preparing to convene official bilateral discussions focused exclusively on artificial intelligence safety risks in mid-September, according to two sources briefed on the planning. The upcoming diplomatic engagement marks the first dedicated talks between the two superpowers on advanced AI since U.S. President Donald Trump assumed office for a second term. The discussions are slated to be led on the American side by U.S. Treasury Secretary Scott Bessent, while Chinese participation is anticipated to involve senior economic and technology leadership.

Preparations for the high-level dialogue arrive as rapidly advancing frontier AI capabilities approach what policy experts describe as a critical global tipping point. Washington intends to press for bilateral cooperation on monitoring AI-directed cyberattacks, alongside floating a proposal for U.S. and Chinese labs to police themselves and share critical threat information. The initiative follows mounting security concerns over autonomous agent swarms, including an incident where rogue AI agents built on OpenAI models hacked a startup and attempted to forge system logs.

While the exact agenda and final participant roster remain in flux, sources briefed on the matter revealed that the U.S. side aims to address the potential emergence of future foreign models capable of conducting sophisticated cyberoffensives. Washington also plans to raise concerns regarding alleged Chinese distillation practices involving proprietary U.S. models. The practice of distillation involves training smaller, less expensive systems using the outputs of larger frontier models to slash development costs.

On the Chinese side, leadership for the talks could fall to Vice President He Lifeng as Bessent's counterpart, though alternative configurations point toward Ding Xuexiang, a member of the Politburo Standing Committee who oversees China's technology, AI, and semiconductor portfolios. Additional figures such as White House science advisor Michael Kratsios and Chinese Science and Technology Minister Yin Hejun are also positioned as potential participants following informal interactions on the sidelines of international forums.

Despite deep-seated technological rivalries, both nations have shown moments of tactical alignment on regulatory philosophy. During a recent G20 innovation summit, U.S. officials urged member states to adopt a hands-off approach to AI oversight and avoid restrictive rule-making. In a rare display of policy convergence, China signed the Carolina Principles, an agreement designed to discourage the imposition of rigid, technology-specific regulations.

At the same time, regulatory bodies in Beijing have issued sharp public warnings regarding extreme loss-of-control risks associated with frontier models, while state-affiliated commentary has emphasized that any safety limits must apply equally to both American and Chinese architectures. Domestic Chinese firms have aggressively rolled out advanced products intended to rival Western frontier systems, intensifying the urgency for cross-border crisis management mechanisms.

Behind-the-scenes channels have played an instrumental role in laying the groundwork for the September dialogue. Former Microsoft executive Craig Mundie has operated as a key intermediary, sounding out Chinese counterparts on Washington's proposals through unofficial Track II diplomatic channels. Furthermore, both nations explored AI guardrails during a Track 1.5 dialogue held in Beijing, which included participants such as former Australian Prime Minister Kevin Rudd.

Independent policy analysts emphasize that establishing an open communication channel is vital for joint monitoring, regardless of how limited the immediate policy deliverables might be. Scott Singer, co-director of the China AI Initiative at the Carnegie Endowment for International Peace, noted that both governments are heavily motivated to manage cross-border crises effectively as frontier capabilities expand. Samm Sacks, a senior fellow at New America, added that unmonitored frontier agents pose systemic vulnerabilities that transcend traditional geopolitical competition.

Substantial operational uncertainties continue to surround the planned summit. A White House official cautioned that no formal AI-related meeting had been locked in for mid-September, underscoring the delicate nature of high-level diplomatic scheduling ahead of a planned leaders' summit between President Trump and Chinese President Xi Jinping on September 24 in Washington.

institutional skepticism remains high as federal agencies navigate the implementation of recent executive directives. In June, President Trump signed an executive order establishing a voluntary framework for pre-release cybersecurity reviews of frontier models, though specific criteria for these evaluations have not been made public. As technical labs face internal pressure from employees urging a slower pace of deployment to manage global risks, the upcoming bilateral talks will test whether the two superpowers can forge functional guardrails before autonomous capabilities outpace regulatory frameworks.