Implementation would hinge on agreed definitions, technical standards, verification mechanisms, and enforcement, noting that the pillars reflect a congressional proposal, not an existing policy.

Stepwise actions include forming joint scientific working groups to develop trusted verification technologies that deter cheating under a treaty.

The approach aims to constrain frontier AI development through diplomacy, dialogue, and potential verification regimes.

The plan treats AI safety as an arms-control-like issue, emphasizing cooperation on risk despite competition, while acknowledging challenges like distrust and definitional disagreements.

A prominent US representative calls for a binding, verifiable U.S.–China AI treaty to prevent loss of human control and mitigate extreme, global risks from advanced AI.

He argues the US and China cannot afford an AI arms race and that keeping humanity in control should be the shared priority guiding any agreement.

The proposal envisions five pillars to govern frontier AI, including pauses on recursive self-improvement and superintelligence, kill switches, air-gapped testing, and robust verification through an international framework.

Five pillars outlined: global pause on recursive self-improvement; global pause on superintelligence; embedded kill switches; mandatory air-gapped testing; and verification with an International AI Inspection Agency.

Additional articulation reiterates the five-pillars framework, detailing the safeguards and the role of an international inspection body.

A Reagan-era trust-but-verify mindset underpins calls for verification technologies and inspection regimes with China.

Congressional actions include a hearing and pressure from Democrats for global action on AI risks ahead of bilateral dialogue.

Domestically, the push includes laws that prevent profit-over-safety in AI and stronger cybersecurity to protect critical model weights, strengthening leverage in negotiations.