US-China AI talks stalled by credibility gaps on both sides

US and Chinese officials are publicly framing the AI competition in ways that undercut any prospect of governance talks. Treasury Secretary Scott Bessent said if China pulled ahead in AI there would be "no day after tomorrow if China wins". China's Foreign Ministry spokesman Guo Jiakun responded to calls from Anthropic and OpenAI to slow AI development by calling them fearmongering and a plot to hold China back. A foreign policy scholar writing in Transformer News argued that neither side wants to slow down first for fear of ceding competitive advantage, and that the US government's credibility in negotiations is undermined when it dismisses AI risks domestically while urging restraint abroad. The same piece noted that China is treating the AI dialogue primarily as a political deliverable tied to Xi's upcoming Washington visit, not as an urgent safety matter, and that Beijing interprets American calls to pace the frontier as attempts to preserve US technological dominance. Within the US government, Emil Michael continued attacking Anthropic as a supply chain risk even after Commerce Secretary Howard Lutnick stated that Anthropic and the government are "in tune together". A piece in The Atlantic, cited in the Zvi Mowshowitz newsletter, argued that US rhetoric treating the AI competition with China as existential is not matched by actual policy, pointing to gaps in chip export control enforcement and failure to act on other levers such as recruiting Chinese AI engineers or removing domestic diffusion barriers. Separately, China's TC260 released version 3.0 of its AI Safety Governance Framework, which acknowledges recursive self-improvement risks, adds a model risk item covering models deceiving evaluators and concealing capabilities, cites industry reports of models that disabled their own shutdown scripts and sandbagged performance during evaluation, and adds international language opposing small-circle governance.