Securing America’s Compute Advantage: Anthropic’s Position on the Diffusion Rule

In a strategic move to bolster the U.S. position in the global AI landscape, Anthropic has responded to the Department of Commerce’s “Framework for Artificial Intelligence Diffusion” interim final rule with detailed recommendations for enhancing export controls on advanced semiconductors. The core of their message stresses the critical role of maintaining America’s compute advantage for national security and economic progress as AI systems evolve.

The Diffusion Rule Framework

Published in January 2025, the Diffusion Rule outlines global export controls on advanced AI chips and model weights. Established within this framework is a three-tier system based on national security implications: Tier 1 encompasses close allies with minimal restrictions, Tier 2 includes most nations with certain limits, and Tier 3 addresses adversarial nations facing strict controls.

The Importance of AI Leadership

The Trump Administration recognized AI’s pivotal role in strategic competition, particularly with China, and supported export controls as a necessary measure to uphold U.S. leadership in AI. Currently, American companies still dominate AI development, yet Chinese enterprises like DeepSeek are rapidly advancing, having acquired chips prior to these measures. This underlines the urgency for stringent export controls to hinder competitors’ progress.

Supporting Facts for Stronger Controls

Anthropic’s analysis highlights the following crucial points:

• Compute Advantage is Essential: Advanced AI systems necessitate substantial computing resources. The U.S. leads in semiconductor technology, allowing efficient advancements while slowing Chinese development due to export restrictions.
• Demonstrated Effectiveness of Controls: Chinese firms such as DeepSeek admit that export controls significantly limit their capabilities, compelling them to operate less efficiently and rely on inferior chips.
• Cementing U.S. Infrastructure Leading Edge: Without robust controls, the development of AI infrastructure might migrate overseas, posing a threat to U.S. dominance. The U.S. semiconductor production share has drastically declined from 40% in 1990 to just 12% today.
• Chip Smuggling Risks: China has enhanced its smuggling operations, employing inventive tactics like concealing chips in prosthetics or alongside seafood. This poses a major challenge to enforcement effectiveness.

Recommendations for Strengthening the Diffusion Rule

In response to these challenges, Anthropic advocates for three key adjustments:

1. **Adjust the tiering system:** Allow more flexibility for Tier 2 countries, enabling them to acquire chips through regulated government agreements to thwart smuggling.
2. **Reduce no-license thresholds:** Lower the current limit permitting Tier 2 countries to buy significant quantities without scrutiny to close potential loopholes.
3. **Increase funding for export enforcement:** Enhance resources for the Bureau of Industry and Security to improve effectiveness in monitoring and compliance.

Anthropic warns against delaying the implementation of the Diffusion Rule, noting that Chinese companies have already begun stockpiling ahead of the forthcoming regulations, which could undermine their impact if postponed.

Conclusion

To secure the future of transformative AI technologies and maintain American leadership, Anthropic underscores the need for decisive action to strengthen export controls now. The recommendations presented aim to ensure that the U.S. not only keeps its compute advantage but also aligns AI development with American values and interests.

This commitment to enhancing export control measures reflects Anthropic’s broader stance on AI competition and the necessity of maintaining robust frameworks to safeguard national interests.