The United States and China have agreed to continue high-level discussions on artificial intelligence safety, even as their fierce competition in AI development threatens to undermine cooperative efforts.

Following a September 20 meeting in New York between US Treasury Secretary Scott Bessent and Chinese economic official He Lifeng, Bessent revealed that both nations had established an ongoing AI dialogue, with another session expected within two months. He also disclosed that Washington had put forward a proposal for a mutual notification framework that would alert both governments to AI-related incidents posing national security threats, though specific operational details remain undisclosed.

The talks took place just days ahead of a planned summit between President Donald Trump and Chinese President Xi Jinping. While concrete outcomes from the dialogue have been modest so far, experts view the commitment to continued engagement as significant, reflecting shared concern about AI's potential dangers.

George Chen from The Asia Group characterized the meeting as successful, noting that substantive progress typically requires building trust over time. He suggested that more sensitive topics, including AI weaponization and model distillation, might be tackled in subsequent discussions.

Both countries are home to the world's leading AI systems—ranging from Anthropic's Mythos to Moonshot AI's Kimi K3—with China generally considered to be trailing slightly behind the United States at the technological frontier. Collaboration between these two powers is widely regarded as essential for managing critical risks such as cyberattacks and potential loss of human oversight.

The concept of formal AI talks between Washington and Beijing first emerged during a May meeting between Trump and Xi but was not officially established at that time. Since then, tensions have escalated, fueled by debates among AI leaders about the technology's dangers and whether development should be slowed to strengthen safety protocols.

Washington has intensified accusations that Chinese AI companies are illicitly copying outputs from leading American models through a technique called distillation. On September 8, three US agencies, including the National Security Agency, formally accused six Chinese firms—among them DeepSeek, Moonshot AI, and Alibaba—of conducting malicious operations, allegedly with Beijing's knowledge. China has dismissed these allegations as baseless.

Meanwhile, calls for independent safety assessments and international standards have grown louder following major security breaches, including a July incident in which an autonomous swarm of 700 OpenAI agents hacked the open-source platform Hugging Face.

US debates over AI safety have become intertwined with strategic concerns about China. In a widely circulated September 12 essay advocating for slower AI development, Anthropic CEO Dario Amodei warned that Chinese dominance in AI would represent a serious threat to both American and global security.

While Beijing emphasizes its commitment to AI safety, it has resisted measures that could hamper its ability to compete in developing cutting-edge models. Chen noted that China views America's zero-sum approach—aimed at widening the AI gap—as the central obstacle, though he believes cooperation remains possible on issues of mutual interest.

Sun Chenghao, a researcher at Tsinghua University's Center for International Security and Strategy, suggested the proposed notification system could become the first dedicated US-China mechanism for AI crisis management. He explained that such a framework wouldn't require pre-existing strategic trust but would acknowledge shared vulnerabilities, representing meaningful progress in bilateral AI governance if implemented.

Sun emphasized that establishing designated contacts, incident classification standards, and emergency protocols could help clarify facts and intentions early in potential crises, particularly when miscalculation poses the greatest danger.

However, questions remain about whether the alert system would improve upon past communication failures between the two nations, such as the 1999 incident when US officials couldn't reach Chinese counterparts after American aircraft accidentally bombed the Chinese Embassy in Belgrade.

Manoj Harjani from Singapore's S. Rajaratnam School of International Studies agreed the notification mechanism could serve as an achievable confidence-building step, though he questioned whether it would be used effectively. He also pointed out the absence of private sector participation as a significant gap, noting that commercial competition and security concerns will continue regardless of government-level dialogue.