US lawmakers propose AI safety framework, hotline with China

US lawmakers Sam Liccardo and Kevin Kiley have proposed legislation seeking binding AI safety safeguards with China, including technical exchanges, model testing, audits and a direct crisis hotline. The framework would also address cybersecurity, CBRN risks and intellectual property concerns.

Washington: Two US lawmakers have proposed legislation seeking binding and verifiable artificial intelligence safety safeguards with China, including a direct crisis hotline between the two governments.

Representatives Sam Liccardo and Kevin Kiley of California said the proposal would promote coordination between Washington and Beijing on risks from advanced, or frontier, artificial intelligence systems.

---

The framework would remove legal barriers preventing American AI safety experts from holding narrowly defined discussions with Chinese counterparts.

It would also direct the State Department and other parts of the administration to negotiate safeguards covering model testing, independent audits, transparency, incident reporting and the verification of compliance.

“A lot of technical work must begin now, before political leadership can ever fruitfully meet at the negotiating table,” Liccardo said.

“Both nations will need to forge shared safety metrics, rigorous evaluation protocols, standardized testing, and critical verification mechanisms,” he added.

The proposal seeks routine technical communication between specialists from the two countries to prevent dangerous AI incidents from escalating into wider crises.

“By establishing red lines now,” Liccardo said, “we can avoid the need for red phones later.”

Under the plan, the United States and China would establish a direct government hotline for urgent communication about serious AI risks or incidents.

Liccardo said engagement with Beijing would be driven by the risks posed by uncontrolled AI development, not by political agreement or trust.

“The AI industry has focused on running faster, but no human wins a race to superintelligence that exceeds human control,” he said.

“We don’t talk to the Chinese because we trust them or agree with them; we shouldn’t and don’t. We talk to the Chinese because—as with the USSR during the Cold War—all of our lives depend on it.”

The legislation would instruct the Commerce Department to pursue protections against the compromise of American intellectual property through model distillation.

Model distillation generally involves transferring capabilities from a larger AI model to a smaller system. US policymakers have raised concerns that the process could be used to reproduce advanced American technology.

The framework would focus negotiations on risks involving the loss of human control over AI systems and failures to keep those systems aligned with their intended purposes.

It would also cover AI-enabled chemical, biological, radiological and nuclear threats, as well as major cybersecurity dangers.

The bill would authorise $100 million for the Center for AI Standards and Innovation. The centre would bring together experts from industry, universities and civil society to develop common testing, evaluation and verification standards.

Participants would be prohibited from sharing proprietary intellectual property, information that could support model distillation, or sensitive cybersecurity and national security information.

The State and Commerce departments would also be required to provide Congress with monthly updates on the negotiations.