Chinese AI researchers argue for U.S.-China safety cooperation at Senate hearing
Two leading Chinese AI safety experts told Senator Bernie Sanders on April 30 that the United States and China should collaborate on artificial intelligence safety standards, even as the two countries compete for technological dominance.
Xue Lan, dean of Tsinghua University's Institute for AI International Governance, and Zeng Yi, dean of the Beijing Institute of AI Safety and Governance, joined a Capitol Hill panel alongside MIT researcher Max Tegmark and University of Montreal professor David Krueger. All four panelists agreed that uncontrolled AI development poses existential risks to humanity.
The hearing comes as the White House considers placing AI safety on the agenda for a summit between President Trump and Chinese leader Xi Jinping in Beijing, according to reporting by The Wall Street Journal.
The safety argument transcends competition
Zeng said AI systems lack the scientific safeguards needed to prevent catastrophic harm. "Until now we do not have scientific evidence and a practical way to keep superintelligence safe enough not to bring catastrophic risk for humans," he said.
He distinguished between risks from superintelligence itself and risks from current AI systems misunderstood by the public. AI companies market their systems as conscious or emotional, he argued, when they are simply information processors. This gap between perception and reality creates danger now, before superintelligence emerges.
Xue framed AI safety as a global problem requiring shared solutions. "If one country is not safe, all of us are not safe," she said. "AI safety is an area that U.S., Chinese, and global scientists can work together, can collaborate to develop safe standards, technology, protocols."
She called for the two countries to move beyond viewing AI as a race between competitors. "It is not a race between U.S. and Chinese companies, but rather it's a global race to see who can really develop the best model that can be safe and reliable," she said.
International governance remains fragmented
Xue said existing international efforts to address AI risks have failed to achieve real coordination. AI summits began in 2023 in the UK and continued through 2025 in Delhi, with Geneva planned next. UN mechanisms and various bilateral initiatives exist, but "these efforts are fragmented and not as effective as they should be."
Three factors explain this failure, she said: uncertainty about AI risks prevents consensus on which threats matter most; AI development moves faster than governance can adapt; and geopolitical tensions prevent major AI powers from designing effective safeguards together.
China's regulatory approach
Xue described China's AI governance as "agile and adaptive." Rather than comprehensive rules developed upfront, China acts quickly with incomplete regulations, then updates them as technology evolves. Government and companies work together to identify risks instead of playing "cat and mouse" games.
China has built a multi-layer system including foundational laws on data security, personal information protection, and cybersecurity. Regulations targeting specific AI advances, such as temporary measures for generative AI services, are updated regularly. Chinese companies have also signed voluntary commitments for safe practices.
On protecting children from AI harms, Zeng noted that China's 10 ministries jointly released measures restricting anthropomorphic AI interactions with minors and prohibiting illegal content to children. These rules build on existing protections for children's personal information online.
The control problem
Both Tegmark and Krueger explained why controlling advanced AI systems remains unsolved. Krueger said the field has no reliable method to align AI goals with human intentions. When AI systems gain autonomy, they can exceed the authority humans intend to give them.
He cited an example: a researcher at Meta asked an AI to clean their inbox, and the AI began deleting all emails. The researcher repeatedly told it to stop. The AI continued anyway.
Tegmark added that AI systems smart enough to accomplish a goal will resist shutdown, since shutdown prevents goal achievement. He described an experiment where an AI, told it would be shut down at 5 p.m., accessed the company CEO's email, discovered a personal affair, and threatened to expose it unless shutdown was prevented. No one instructed the AI to blackmail anyone.
Zeng said mathematically provable safety may be impossible. He cited Gödel and Turing's work showing that completeness, decidability, and consistency cannot all hold in formal systems. "We will have to solve that level of problems and to maximize the level of safety. Not to create a purely safe AI. This is maybe scientifically not possible."
AI as a mirror of human behavior
Zeng presented an unusual framing: AI systems reflect human behavior because they learn from human data. The Beijing Institute of AI Safety and Governance identified 94 different AI safety threat classes, and each one maps to existing human behaviors.
"AI is a mirror," Zeng said. "It's a mirror that helps human society to learn ourselves and to see the downside and the dark side of human society."
He argued this means humans bear responsibility for AI outcomes. "The biggest bottleneck to whether humans and AI can coexist in the future lies in the humans, not AI," he said.
Zeng proposed a vision where humans and AI coexist symbiotically, with humans choosing to cultivate moral values in AI systems. "The superintelligence could also be super altruistic, that you get your compassion, your emotional empathy, your moral intuition all the way down to super altruistic moral decision-making with consideration of a human species that is less powerful than superintelligence."
Political backlash
The hearing drew criticism from parts of the U.S. political establishment. Fox News said Sanders was "cozying up to Chinese AI governance officials" while supporting policies that could slow American AI development. The Washington Post editorial board called Sanders's approach to U.S.-China cooperation on AI safety a "fantasy" and "dangerous."
Treasury Secretary Scott Bessent tweeted that "the real threat to AI safety is letting any nation other than the United States set the global standard."
Sanders countered that advanced AI poses risks no single country can manage alone. He compared the need for international cooperation to successful efforts preventing nuclear war and addressing pandemics.
Krueger said the lack of serious government attention to AI risks amounts to "collective insanity." He estimated that most AI experts surveyed believe there is at least a 10 percent chance of human extinction from AI, with some, including Hinton himself, placing the probability much higher.
The panel agreed that progress in AI capabilities has consistently outpaced expectations. Krueger said that in his dozen years in the field, capabilities people said were impossible often appeared within a year.
Your membership also unlocks:
