carnegieendowment.org via Reddit

China Closed AI Gap Despite Four Years of Strict Rules

regulation china ai policy us-china-ai

Key insights

  • China enforced binding AI regulations since 2021, including 90 percent accuracy requirements for thousands of sensitive questions, yet still closed the technological gap with Silicon Valley.
  • DeepSeek's V3 release in early 2025 shook Silicon Valley's confidence despite operating under China's extensive and burdensome regulatory environment.
  • Sheehan argues capital, research talent, and computing power determine AI competitive outcomes, not the permissiveness of a country's regulatory regime.

Why this matters

Trump's EO sets a thirty-day pre-release testing window for powerful AI models, but Sheehan's analysis directly challenges the policy's core rationale that deregulation produces competitive advantage against China. For founders and technical leaders, this reframes where resources should flow: compliance burden reduction is not the bottleneck when capital, research talent, and computing power are the actual determinants of AI outcomes. The risk is that framing the US-China AI race as a deregulation contest leads policymakers to dismantle technically informed oversight at bodies like the Commerce Department's Center for AI Standards and Innovation, while leaving compute access and capital formation, the variables that actually matter, unaddressed.

Summary

Trump's AI executive order requires government testing of new models thirty days before public release, down from a ninety-day version he canceled. Trump's justification: "We're leading China...and I don't want to do anything that's going to get in the way." Carnegie senior fellow Matt Sheehan argues that framing is backward. China enforced the world's most extensive AI regulations for four years, including a 2023 rule requiring 90 percent accuracy on thousands of sensitive political questions. DeepSeek V3 still shook Silicon Valley in early 2025. Essentially: (Anthropic, DeepSeek) make the regulation-competitiveness link hard to prove, both thriving under opposite regulatory regimes. - China's binding rules span recommendation algorithms, deepfakes, and generative AI since 2021. - Sheehan calls China's approach "small, fast, flexible" with direct government-company engagement building regulator expertise. - Capital, research talent, and computing power drive AI outcomes; regulatory permissiveness doesn't create them. The U.S. holds structural advantages in capital and compute. Deregulation alone won't expand either.

Potential risks and opportunities

Risks

  • Anthropic and other frontier labs operating under the thirty-day window may face competitive pressure to race through government testing, degrading the quality of cybersecurity evaluations the EO depends on
  • If deregulation is treated as the primary competitive lever, the Commerce Department's Center for AI Standards and Innovation could lose the direct-engagement model Sheehan identifies as a genuine U.S. strength
  • Chinese AI companies already benefiting from a 'small, fast, flexible' regime while the U.S. reduces structured oversight could widen the regulatory expertise gap between the two countries' agencies

Opportunities

  • The Commerce Department's Center for AI Standards and Innovation is positioned to become the benchmark for technically informed AI engagement if it can operationalize Sheehan's direct-company model at scale
  • AI safety and pre-release testing vendors gain a newly mandated government customer base under the executive order's thirty-day requirement for powerful new models
  • Anthropic, with Mythos already cited for 'extraordinary cybersecurity capabilities' in the EO's context, is positioned to shape what government testing standards actually measure and how they are scoped

What we don't know yet

  • Whether the thirty-day pre-release testing window applies to open-weight models like DeepSeek V3, which the article does not specify
  • How Anthropic's Mythos model's 'extraordinary cybersecurity capabilities' were evaluated, and by whom, before being cited in the executive order's framing
  • Whether the Commerce Department's Center for AI Standards and Innovation has the staffing and budget to scale direct company engagement at the pace Sheehan recommends

Originally reported by carnegieendowment.org

Read the original article →

Original headline: Carnegie Endowment: Trump's AI Order Won't Stymie US-China Competition — China Regulated AI for Four Years and Still Closed the Gap With Silicon Valley