IFP Think Tank Publishes 23 Policy Recommendations to Govern Automated AI R&D Risks
Summary
- • Institute for Progress (IFP) published 23 specific policy ideas for managing risks of automated AI R&D
- • Key proposals include funding CAISI at $84M/year and creating an AI Verification Consortium (AIVEC)
- • Report recommends strengthening chip export controls and protecting model weights from adversarial theft
- • IFP frames the paper as helping policymakers address risks before AI meaningfully automates its own improvement
Details
Transparency Mandates (Recs. 1-2)
Frontier AI companies should publicly disclose AI R&D automation trends and risks; Congress should mandate incident reporting and whistleblower protections
CAISI Budget Recommendation
IFP recommends Congress fund the Center for AI Standards and Innovation (CAISI) at a minimum of $84 million per year (Rec. 3)
Agency Role Clarity (Rec. 4)
The White House should assign clear roles and responsibilities across US government agencies to increase AI policy specialization
Intelligence Collection (Rec. 5)
Intelligence agencies should improve collection and analysis on foreign AI development and counter threats targeting US AI companies
AI Verification Consortium (AIVEC)
CAISI should co-lead AIVEC with industry to prototype compute monitoring, hardware testbeds, and AI verification technologies (Recs. 7-11)
Verifiable Data Center
AIVEC should coordinate construction of a fully verifiable data center as a proof-of-concept for AI transparency and oversight (Rec. 10)
DARPA/NSF Verification R&D
DARPA and NSF should establish dedicated AI verification R&D programs (Rec. 11)
Chip Export Controls (Recs. 15-16, 18)
Congress and BIS should strengthen semiconductor controls, close gaps in AI chip exports, and maintain visibility into overseas US chip sales
Counter Adversarial Distillation (Rec. 17)
FTC, DOJ, BIS, CAISI, and Congress should help industry counter adversarial distillation of US AI model capabilities by foreign rivals
Protect Model Weights (Rec. 19)
Government and FFRDCs should establish consensus security guidelines for protecting model weights from theft, prototyped in a government facility
Energy and Construction (Recs. 20-21)
Congress should ensure sufficient electrical capacity and enable domestic data center construction to sustain US AI leadership
Source: Institute for Progress (IFP) policy paper, reported via Import AI newsletter. Recommendations numbered per original report.
What This Means
IFP's 23-recommendation framework is one of the most operationally specific US policy roadmaps yet published for governing the risks of automated AI research and development—the scenario where AI systems meaningfully accelerate their own improvement. Unlike aspirational frameworks, it assigns concrete budgets, institutions, and program designs, making it actionable for policymakers. The report signals growing think-tank consensus that AI R&D automation risk is no longer theoretical, and that the US needs dedicated verification infrastructure—not just regulatory language—to stay ahead of both safety risks and geopolitical competition.