In response to the White House’s Request for Information on an AI Action Plan, Anthropic has provided a set of recommendations to the Office of Science and Technology Policy (OSTP).
These recommendations are designed to position the United States to capitalize on the forthcoming economic and national security advantages that will arise with more powerful AI systems. Anthropic projects that advanced AI systems will arrive around late 2026 or early 2027, and will demonstrate capabilities that match or exceed those of Nobel Prize winners across various disciplines including biology, computer science, mathematics, and engineering. These systems will be able to interface with all digital tools a human uses today, including text, audio, and video. They will control technology autonomously and reason independently through complex tasks for extended periods.
Anthropic’s most recent work, including the release of Claude 3.7 Sonnet and Claude Code, demonstrates the rapid advancement of AI capabilities. The company urges the U.S. to take decisive action to maintain technological leadership, focusing its recommendations on six critical areas:
Key Recommendations
-
National Security Testing: The government should develop strong evaluation capabilities, creating testing frameworks, secure infrastructure, and expert teams to analyze potential vulnerabilities in deployed AI systems.
-
Strengthening Export Controls: Anthropic advocates for stricter regulations that allow America and its allies to benefit from the potential of powerful AI systems, making it difficult for adversaries to have access to the AI-empowering infrastructure. Restrictions should include controlling H20 chips, requiring government agreements for countries deploying chips, and tightening no-license-required thresholds.
-
Enhancing Lab Security: As AI systems become strategic assets, classified communication channels should be established between AI labs and intelligence agencies. This will promote accelerated security clearances for industry professionals and promote development for next-generation infrastructure security standards.
-
Scaling Energy Infrastructure: Anthropic suggests setting an ambitious goal to build an additional 50 gigawatts of power by 2027 to lead AI development, streamlining permitting, and improving approval processes.
-
Accelerating Government AI Adoption: The government should take inventory of workflows that could benefit from AI support. Agency leaders should deliver programs where AI can generate significant public benefit.
-
Preparing for Economic Impacts: To ensure that the AI benefits are shared within society as a whole, the government should modernize data collection methods and prepare for potential large-scale changes to the economy.
These recommendations build on Anthropic’s previous policy work. Anthropic seeks to enable innovation while minimizing the inherent risks associated with increasingly capable AI systems.
