Chinese AI Companies and EU/U.S. Frontier Safety Regulation

Regulators in the European Union, California, and New York have enacted legislation requiring developers of advanced AI models to disclose or implement safety practices to address catastrophic risk, including capability evaluations, risk assessments, and incident reporting. While these frameworks have extraterritorial application, their implications for Chinese AI companies-which have achieved significant international attention through open-source distribution and consumer applications-have not been closely studied. In this paper, we systematically analyze the legal obligations Chinese AI developers face under the EU AI Act, California's SB 53, and New York's RAISE Act. We map compute thresholds, market placement definitions, open-source exemptions, and enforcement mechanisms to the current Chinese AI landscape, examining their applicability to companies including DeepSeek, Alibaba, ByteDance, and Z.ai. Under the EU AI Act, Chinese frontier models clearly qualify as general-purpose AI models subject to baseline Article 53 obligations (transparency, copyright compliance, documentation), though whether specific models qualify for open-source exemptions depends on license terms, as some Chinese models use restrictive licenses that disqualify them. Regarding heightened systemic risk requirements under Article 55, most Chinese models fall below the 10^25 FLOP threshold, though some models from ByteDance and Z.ai may exceed it, and the Commission may designate compute-efficient models like DeepSeek based on equivalent capabilities. U.S. state legislation applies a higher 10^26 FLOP threshold that current Chinese models do not reach. Enforcement potential differs across companies: DeepSeek maintains no disclosed EU or U.S. corporate presence, while Alibaba and ByteDance have assets that could serve as enforcement points. Due to interest by Chinese AI companies in EU and U.S. consumer markets, regulation in these jurisdictions may provide a noteworthy incentive for Chinese AI companies to implement frontier safety practices.