US and China Urged to Forge AI Treaty Amid Existential Risk Concerns

Representative Ro Khanna has advocated for a binding, verifiable U.S.-China AI treaty to avert a catastrophic arms race in artificial intelligence, emphasizing that humanity stands to lose if control of the technology is relinquished.

Khanna, the ranking member of the House Select Committee on China, recently convened the first congressional hearing focused on the need for such an agreement. This initiative, coupled with a letter from Democrats on the committee to the White House, preceded an announcement by President Trump and Chinese General Secretary Xi of a bilateral dialogue on AI, a development Khanna credits to congressional pressure.

Beyond discussion, Khanna proposes the formation of joint technical working groups comprising top scientists from both nations. These groups would develop mutually trusted verification technologies to ensure compliance with a future AI treaty. "This is an easy second step we should take today," Khanna stated.

Pillars of a Proposed AI Treaty

Khanna outlined five key pillars for a treaty aimed at safeguarding humanity from unchecked AI development:

• Pause on Recursive Self-Improvement: A worldwide halt on AI developing AI without human intervention, identified as the most probable path to losing control of the technology.
• Pause on Superintelligence: A global moratorium on the development of AI that surpasses human capabilities across all domains, which could have unpredictable and potentially dangerous consequences.
• Embedded Kill Switches: The integration of kill switches in all advanced AI models to provide a mechanism for intervention if rogue AI systems pose a threat.
• Air-Gapped Testing: Mandatory testing of all AI models in isolated, air-gapped environments before release to prevent catastrophic misuse, whether by criminal actors, terrorists, or misaligned AI agents.
• Verification and Transparency: Adherence to President Reagan's principle of "trust, but verify," especially concerning China. This includes establishing verification technologies and inspection regimes, potentially through an International AI Inspection Agency, to ensure Beijing cannot circumvent agreements and gain an unfair technological advantage.

Khanna drew parallels to the Cold War, noting that while the U.S. and Soviet Union were adversaries, leaders like Reagan and Gorbachev recognized the existential threat of nuclear war. He asserts that an AI war would be similarly catastrophic for civilization.

Even as the U.S. holds China accountable for economic practices and human rights abuses, Khanna believes both nations can unite around the common goal of maintaining human control over AI. A global pandemic unleashed by rogue AI, he argues, would not discriminate between nationalities, making cooperation essential for the survival of humankind.

Khanna stressed that while a verifiable AI treaty will require significant effort, delaying these negotiations could lead to severe consequences. In parallel, he urged domestic action, including legislative measures to prevent AI companies from prioritizing profit over humanity's collective fate and strengthening cybersecurity within American AI labs to prevent intellectual property theft by China.

He concluded by stating that confronting AI risks requires a unified approach, transcending partisan divides. The advent of superintelligence, he suggested, represents a pivotal moment for the human species, potentially surpassing the impact of the Industrial Revolution or the development of the atomic bomb. Khanna called for collective action to ensure humanity remains in control of its technological future.