RAND recommends the U.S. preserve options in an uncertain path to superintelligence

The RAND Corporation has published a lengthy paper arguing that, because the trajectory toward superintelligent AI is deeply uncertain, the United States’ best approach is to preserve strategic flexibility—a "Freedom of Action" strategy. RAND’s central claim is that, given the many unknowns about how the next phase of AI development will unfold, the U.S. should invest now in capabilities that keep future policy and operational options open.

Objective of the strategy

The stated purpose of the Freedom of Action strategy is "to build and preserve the ability to secure U.S. geopolitical advantage and ensure humanity’s survival with agency through the transition to Superintelligence."

Four building blocks of the strategy

1. Build a human–AI ecosystem

• Invest in AI safety and develop tools that preserve human agency.
• Invest in mechanisms to share the benefits of AI and prepare society for AI-driven disruption.
• Shape incentives within the AI ecosystem.

2. Build an AI-security architecture

• Develop tools to understand frontier AI systems domestically and abroad, including mechanisms to gain visibility into underlying compute.
• Create technologies to verify potential agreements.
• Build regulatory expertise to govern the technology.

3. Adapt legacy national security enterprises for the AI era

• Overhaul national security institutions so they are ready for AI-driven changes and can capitalize on AI capabilities.

4. Invest in the capacity of citizens, firms, and governments to respond

• Provide decisionmakers with information to think clearly about AI risks and tradeoffs.
• Develop and diffuse beneficial uses of AI to help manage change.
• Create "break-glass" contingency plans for crises and raise AI literacy.

Seven archetypal national strategies in three families

RAND defines seven archetypal strategies, grouped in three families:

Coexistence:

• Dominance: the U.S. leads AI development and suppresses rivals.
• Co-Development: the U.S. leads a consortium (including China) to co-develop safe superintelligence with shared governance, pooled compute, and verification and monitoring.
• Preparedness: the U.S. mobilizes a coexistence approach through informal coordination, voluntary commitments, phased deployment, independent evaluation, incident reporting, and domestic capacity to absorb and recover from AI incidents.

Denial:

• Moratorium: pursue a verifiable global halt to AI development above specified thresholds.
• Deterrence: the U.S. halts its own frontier development above defined thresholds and uses the full toolkit of national power to prevent or, if necessary, destroy foreign programs.
• Continuity of Society: if preventive strategies fail or are infeasible and coexistence is judged impossible, the remaining objective is survival with some human populations retaining agency—creating geographically distributed, biologically self-sustaining settlements hardened against AI-enabled threats.

Acceleration:

• This family assumes that constraining AI development may be more dangerous than advancing it. It relies on markets, competition, and rapid iteration, hoping that safety will emerge as a byproduct of useful AI.

Five main uncertainties that shape strategy choice

RAND identifies five key uncertainties that determine which strategy might be optimal:

• Danger proximity: how close is AI to becoming so dangerous that risks outweigh benefits? If danger is close, buy-time and resilience strategies are preferred; if not, accelerated development may be warranted.

• Coexistence feasibility: can humans and advanced AI find a stable equilibrium? If feasible, strategies that build and deploy advanced AI gain appeal; if infeasible, the case shifts toward restraint or suppression.

• Restraint feasibility: can actors cooperate to restrain AI development? If yes, cooperative strategies are available; if not, restraint requires coercive enforcement.

• Decisive strategic advantage: can a single actor achieve a decisive lead? If so and if the U.S. can sustain dominance, unilateral strategies may make sense; if not, no single actor can dictate global terms.

• Suppression feasibility: can domestic or rival programs be suppressed via technological controls? If suppression is feasible, Deterrence becomes viable; if not, actors can press ahead without fear of being restrained.

Why this matters — taking AI seriously requires spending money

RAND’s principal practical takeaway is that the United States must spend significant resources now if it wants to preserve optionality and pre-position to benefit from continued AI progress. Even if the main objective is merely to keep options open, doing so requires investment. RAND contrasts this recommended posture with what it sees as a prevailing U.S. tendency toward "acceleration": prioritizing faster capability development and competition while comparatively underinvesting in proactive safety, preparedness, and institutional adaptation.

Summary

RAND recommends a cautious but proactive posture: because the path to superintelligence is uncertain, the U.S. should build capabilities that preserve freedom of action across multiple plausible futures. That approach entails sustained investment across human–AI ecosystems, security architecture, national security adaptation, and societal readiness, backed by an appreciation of the strategic uncertainties that will determine which concrete policies make sense.