Anthropic Submits Recommendations to US OSTP for AI Action Plan

Anthropic submitted recommendations to the OSTP (Office of Science and Technology Policy) in response to the White House's Request for Information (RFI) on the AI Action Plan. Based on the performance improvements of Claude 3.7 Sonnet, Anthropic forecasts that powerful AI systems with Nobel Prize winner-level intellectual capabilities will emerge by late 2026 or early 2027.

These systems are expected to have the ability to autonomously control not only digital tasks but also the physical world, such as experimental equipment or robotic systems. Accordingly, Anthropic proposed 6 key areas that the US should focus on to maintain its technological leadership.

• National Security Testing: Building standardized evaluation frameworks and testing infrastructure to analyze security vulnerabilities in domestic and foreign AI models.
• Strengthening Export Controls: Tightening semiconductor export restrictions to prevent adversarial nations from accessing AI infrastructure, including H20 chip controls and pursuing government-to-government agreements.
• Enhancing Lab Security: Establishing confidential communication channels between AI labs and intelligence agencies and streamlining security clearance procedures.
• Expanding Energy Infrastructure: Securing an additional 50GW of dedicated power by 2027 and simplifying permitting procedures.
• Accelerating Government AI Adoption: Conducting a comprehensive survey of government workflows that could utilize AI and advancing programs for the public interest.
• Preparing for Economic Impact: Modernizing economic data collection mechanisms to prepare for large-scale economic shifts caused by AI.