Response to OSTP RFI: Items to Include in the Trump 2025 AI Action Plan

The Center for AI Policy (CAIP) responded to the White House Office of Science and Technology Policy (OSTP) request for information to provide input on President Trump's forthcoming 2025 AI Action Plan.

Executive Summary

Thank you for the opportunity to provide input on President Trump’s 2025 AI Action Plan. The Center for AI Policy (CAIP) strongly agrees with the White House’s assessment that “with the right government policies, the United States can solidify its position as the leader in AI and secure a brighter future for all Americans.”

Unfortunately, we are not currently on a path to that brighter future. Today’s AI is fundamentally insecure and unreliable. Advanced AI models produce incorrect outputs, deceive their creators, and refuse human orders to shut down. AI is also increasingly able and willing to coach arbitrary users – including terrorists and criminals – on how to launch automated cyberattacks and design novel bioweapons.

As these models grow more powerful, the consequences of blindly trusting them will grow more severe. AI is likely to direct most of our weapons, our energy grid, and our communications. In this context, even a moderate AI failure could result in losing access to critical infrastructure. According to a former OpenAI researcher, “in the extreme, failures could look more like a robot rebellion.” A total loss of control over AI could quickly disenfranchise, harm, and kill millions of people.

Bold national leadership will be needed to avoid these negative consequences. To ensure that American AI promotes human flourishing and national security, CAIP urges the Trump Administration to introduce third-party national security audits for advanced AI. These audits will enable security agencies to better understand the threat landscape and verify the security of America’s leading technologies. CAIP views these audits as the number one priority for the Trump Administration’s AI policy.

To further strengthen American leadership in AI, CAIP also recommends the following initiatives:

1. Gather cyber incident data by recognizing frontier AI as essential infrastructure
2. Accelerate the National Science Foundation’s AI explainability research
3. Hire more field agents to enforce export controls on advanced AI chips
4. Insist that frontier AI developers participate in emergency response planning
5. Protect U.S. civilians against drone attack

Read the full response here.