Khanna Demands Answers from Chinese AI Labs, Intelligence Community on Catastrophic AI Risks

WASHINGTON – Following a first-of-its-kind congressional hearing on the need for a binding, enforceable, and verifiable U.S.-China agreement to address catastrophic artificial intelligence risks, Ranking Member Ro Khanna (D-CA) of the House Select Committee on the Strategic Competition Between the United States and the Chinese Communist Party is pressing leading Chinese AI companies and the U.S. Intelligence Community for greater transparency and preparedness as frontier AI capabilities rapidly advance.

In letters sent to DeepSeek CEO Liang Wenfeng, Alibaba CEO Eddie Wu, and Moonshot AI CEO Yang Zhilin, as well as Director of National Intelligence Jay Clayton, Khanna called for immediate action to address the risks of humans losing control of advanced AI models and urged the development of the safeguards, monitoring capabilities, and verification mechanisms necessary to prevent catastrophic misuse.

“Neither the U.S. nor China wins if humans lose control of this technology or misaligned AI triggers a global economic, biological, or nuclear crisis,” Khanna wrote to the Chinese AI executives.

The letters follow a September 23 congressional hearing convened by Khanna that brought together leading experts to examine catastrophic AI risks, including loss of control, misalignment, and the potential misuse of advanced AI systems to facilitate biological, chemical, nuclear, or cyber harms. At the hearing, Dr. Geoffrey Hinton and Lt. General Jack Shanahan testified that they do not believe the U.S. government currently has sufficient capabilities to monitor advanced AI models at frontier laboratories.

Khanna previously called on DeepSeek, Alibaba, and Moonshot AI to join leading American AI laboratories in a commitment to “pace the frontier” by slowing the development of increasingly advanced AI systems until appropriate societal safeguards can catch up. The companies did not respond by the September 22 deadline.

As Khanna wrote, “this hardly surprise[s] me given PRC AI companies’ complicity in grave human rights abuses in China, censorship in defiance of the principles of freedom of expression that belong to all of mankind, support for the People’s Liberation Army and the PRC’s intelligence services, and pattern of illicitly distilling U.S. models in defiance of American companies’ terms of use…[W]ith every refusal by your companies to participate in this effort to seek a verifiable global slowdown of AI development to benefit all of humanity, you only strengthen the case for a binding and enforceable AI treaty between our governments.”

In supplemental letters, Khanna is now demanding detailed information from the three Chinese AI companies about their development and governance of advanced AI, including:

• Recursive self-improvement: How the companies are developing or planning to develop AI systems capable of recursively improving themselves and what safeguards are in place against loss-of-control scenarios.
• Superintelligence and AGI: Their plans and timelines for developing artificial general intelligence or superintelligence and safeguards against associated catastrophic risks.
• Kill switches: Efforts to develop and implement mechanisms that could disable highly capable AI systems in the event of a loss of control scenario.
• Auditing and testing: How advanced models are tested before release, including whether they are evaluated in air-gapped environments and screened for capabilities that threaten all of humanity, such as the development of dangerous biological capabilities.
• Verification: Whether Chinese AI laboratories will participate in U.S.-China technical working groups exploring technologies capable of verifying compliance with a future international AI agreement and whether they would permit independent inspections.

These five lines of questioning align with the five pillars of a U.S.-China AI treaty Khanna laid out in a letter to Secretary of State Rubio.

“Simply put, it is clear we cannot gamble the future of humanity on whether AI companies – including Chinese companies like yours – will responsibly self-regulate,” Khanna wrote.

At the same time, Khanna is calling on the U.S. Intelligence Community to strengthen the government’s ability to understand and respond to these risks. In his letter to DNI Jay Clayton, Khanna requested two formal assessments from the Intelligence Community.

First, Khanna requested an assessment of how effectively the U.S. government could respond to a loss-of-control scenario involving advanced AI, including the government protocols and capabilities that would be needed if a catastrophic scenario emerged.

Second, he requested an intelligence assessment of General Secretary Xi Jinping’s and the Chinese Communist Party’s understanding of and approach to catastrophic AI risks, including loss of control, misalignment, and the potential use of advanced AI to facilitate biological, chemical, or nuclear harm.

Khanna also urged ODNI to direct its technical and analytical resources toward developing the capabilities necessary to effectively monitor PRC frontier AI laboratories. At last week’s hearing, leading AI experts testified the U.S. Government currently lacks sufficient resources to monitor frontier labs generally.

“The Intelligence Community also has an important role to play in helping ensure an eventual treaty is complied with,” Khanna wrote.

Furthermore, he has called for the establishment of joint U.S.-China technical working groups composed of leading scientists to develop and establish mutual trust in verification technologies that could underpin an eventual international AI treaty.

“History will look back at this moment, the choices made by humanity as we potentially approach the Rubicon of superintelligence or AGI, and either celebrate our foresight or condemn our inaction,” Khanna concluded. “Our leaders, including the membership of my Committee, need to be informed of our government’s readiness for these forthcoming risks and the PRC’s approach to them.”

Read the letter to DeepSeek, Alibaba, and Moonshot AI HERE.                                  

Read the letter to ODNI HERE.

###