The White House unveiled "Winning the Race: America's AI Action Plan," a comprehensive blueprint for maintaining US dominance in artificial intelligence development. Anthropic welcomes the plan's emphasis on expanding AI infrastructure and government implementation, alongside enhanced safety protocols and security measures. Numerous elements in the plan align with Anthropic's recommendations submitted to the Office of Science and Technology Policy (OSTP) earlier. However, rigorous export restrictions and AI development transparency requirements remain essential components for safeguarding American AI superiority.
Expanding AI Infrastructure and Implementation
The Action Plan makes AI infrastructure and government adoption a top priority, mirroring Anthropic's OSTP submission from March.
Anthropic endorses the Administration's pledge to expedite data center and energy infrastructure approvals to meet AI's growing power demands. As outlined in Anthropic's OSTP submission and Pennsylvania Energy and Innovation Summit presentation, insufficient domestic energy resources could compel American AI companies to move operations abroad, risking technology exposure to hostile nations. Anthropic's "Build AI in America" report provides detailed recommendations for the Administration to expand national AI infrastructure capabilities.
The Plan's federal AI adoption proposals closely match Anthropic's policy objectives and White House recommendations, encompassing:
- Directing OMB to resolve resource limitations, procurement challenges, and operational barriers hindering federal AI implementation
- Initiating an RFI to pinpoint federal regulations constraining AI advancement, with OMB leading reform initiatives
- Modernizing federal acquisition protocols to eliminate obstacles preventing agency AI deployment
- Advancing AI integration in defense and national security through collaborative public-private efforts
Expanding AI Access and Benefits
Anthropic supports the Action Plan's commitment to ensuring widespread engagement in and advantages from AI's ongoing evolution and implementation.
The continuation of the National AI Research Resource (NAIRR) pilot program guarantees that students and researchers nationwide can engage with and advance AI technology. Anthropic has consistently backed the NAIRR and maintains a partnership with the pilot initiative. The Action Plan's focus on accelerated worker retraining programs and preliminary AI apprenticeships acknowledges past technological transition failures and shows dedication to distributing AI advantages equitably.
Supporting these initiatives are Anthropic's studies on AI's economic transformation. The Economic Index and Economic Futures Program provide researchers and decision-makers with essential data and resources to ensure widespread distribution of AI's economic advantages while properly addressing potential risks.
Advancing Secure AI Innovation
Advanced AI systems will emerge in the near future. The plan's focus on preventing powerful AI model misuse and addressing forthcoming AI-related challenges is both fitting and commendable. Anthropic particularly supports the administration's commitment to funding research in AI interpretability, control mechanisms, and adversarial resilience - critical research areas for managing powerful AI systems.
Anthropic appreciates the Action Plan's endorsement of NIST's Center for AI Standards and Innovation (CAISI) in assessing frontier models for national security implications and anticipates continuing collaboration. The Administration should maintain investment in CAISI, as sophisticated AI systems demonstrate alarming progress in biological weapons development capabilities. CAISI leads in creating testing and assessment methods to mitigate these dangers, which should concentrate on the most distinct and severe national security concerns.
Establishing National Standards
Beyond evaluation, Anthropic believes fundamental AI development transparency measures, including public disclosure of safety assessments and capability evaluations, are crucial for responsible AI progress. Top AI model creators should meet basic, publicly-confirmable standards for evaluating and controlling catastrophic risks from their systems. Anthropic's suggested frontier model transparency framework addresses these concerns, though the report could have emphasized this area more strongly.
Major laboratories, such as Anthropic, OpenAI, and Google DeepMind, have already adopted voluntary safety protocols, proving that responsible practices and innovation can thrive together. With Claude Opus 4's release, Anthropic preemptively implemented ASL-3 safeguards against CBRN weapons development misuse, demonstrating that comprehensive safety measures enhance system reliability and quality.
Anthropic agrees with the Administration's worries about excessive regulatory approaches producing an uneven and cumbersome legal landscape. These transparency standards should ideally originate from federal authorities through unified national requirements. Nevertheless, consistent with Anthropic's position that a decade-long state AI law freeze is excessive, opposition continues against measures preventing states from safeguarding citizens against potential powerful AI system harms when federal action is absent.
Preserving Robust Export Restrictions
The Action Plan identifies preventing adversaries from accessing advanced AI computing as crucial for both strategic competition and national defense. Anthropic firmly concurs, making the Administration's recent H20 chip export reversal to China particularly troubling.
AI advancement follows scaling principles: system intelligence and capabilities correlate with training compute, energy, and data scale. While these principles persist, cutting-edge reasoning models show AI capability also scales with inference compute availability for specific tasks. Inference compute is constrained by chip memory bandwidth. Although Huawei chips surpass the H20's processing power, production challenges persist and no Chinese-manufactured chip equals the H20's memory bandwidth capabilities.
Consequently, the H20 delivers essential computing features unavailable to Chinese entities otherwise, compensating for China's significant AI chip deficit. Permitting H20 exports to China would waste the chance to extend American AI supremacy during this new competitive phase. Additionally, US AI chip exports won't deter China's pursuit of AI stack independence.
Anthropic strongly advocates maintaining H20 chip restrictions, aligning with the Action Plan's export control recommendations and essential for preserving America's AI advantage.
Future Outlook
The correspondence between Anthropic's suggestions and the AI Action Plan reflects mutual recognition of AI's revolutionary possibilities and necessary steps for sustaining American preeminence.
Anthropic anticipates collaborating with the Administration on implementing these programs while maintaining focus on catastrophic risk prevention and robust export controls. This partnership can guarantee powerful AI systems develop safely within America, by American enterprises, embodying American principles and priorities.
For comprehensive policy recommendations, consult Anthropic's complete OSTP submission, responsible AI development efforts, and recent domestic energy capacity expansion report.