US, China Are Discussing AI Guardrails to Safeguard Most Powerful Models, Bessent Says

U.S. and Chinese delegations are engaged in discussions about establishing AI guardrails during a Beijing summit, aiming to create protocols that prevent non-state actors—including criminals and terrorists—from exploiting the most advanced AI models. Treasury Secretary Scott Bessent stated in a CNBC interview that maintaining U.S. leadership in AI is a priority, though the focus remains on fostering innovation without stifling progress. He described the goal as achieving the ‘highest performance calculus’ to maximize both innovation and safety. The urgency of these talks follows the release of Anthropic’s Mythos AI tool, which has uncovered significant software security flaws, prompting banks and financial institutions to urgently patch vulnerabilities in their systems. Government officials have warned that such tools could be weaponized to disrupt global markets and financial stability. Bessent confirmed that the U.S. Treasury is coordinating with the 11 largest banks to address these risks and will extend similar efforts to smaller regional and community banks. Bessent also highlighted that the U.S. is consulting with major AI developers, including Alphabet’s Google and OpenAI, calling them ‘very good partners.’ He noted that future advancements in AI models will likely introduce similar ‘step function’ improvements, necessitating proactive measures. The Treasury has assured a ‘smooth transition’ to these new capabilities while emphasizing that discussions with China on AI would not occur if China were leading the technology race. The U.S. approach prioritizes integrating American best practices and values into AI governance frameworks, which will then be promoted globally. Meanwhile, Nvidia CEO Jensen Huang attended the summit in Beijing, where Reuters reported that the U.S. has approved around 10 Chinese companies to purchase Nvidia’s H200 chip, a critical component for AI infrastructure. Bessent clarified that such approvals fall under the Commerce Department’s jurisdiction, not Treasury oversight. The talks underscore the geopolitical tension between fostering collaboration and safeguarding technological dominance. While China’s interest in guardrails signals a willingness to engage, the U.S. remains cautious about sharing control over AI development, particularly as vulnerabilities like those exposed by Mythos pose growing risks to global security.