RAND Superintelligence Plan Pushes U.S. AI Options Beyond Acceleration

A RAND document on superintelligence strategy gives the United States a deliberately flexible policy brief, with Import AI highlighting the central recommendation: spend now to keep multiple national AI options open rather than betting on one path.

The proposed “Freedom of Action” strategy is not a single moratorium, race plan or alliance design.

RAND frames the goal as preserving U.S. advantage and human agency through a possible transition to superintelligence.

That approach treats the next phase of AI development as uncertain enough that policy capacity itself becomes the first asset to build.

RAND’s first ingredient is a human-AI ecosystem.

The document calls for investment in AI safety, tools that preserve human agency, mechanisms to share benefits, preparation for AI-driven disruption and incentives that shape the behavior of the AI ecosystem.

The second ingredient is an AI-security architecture: visibility into advanced systems and the compute behind them, technology for verifying possible agreements and regulatory expertise able to govern frontier systems.

The strategy then moves from technology governance into state capacity.

Legacy national security institutions would need to be adapted for the AI era so they can use AI capabilities while handling the risks those capabilities create.

Citizens, companies and governments would also need response capacity, including information for decision-makers, positive uses of AI that help society adjust, break-glass crisis plans and broader AI literacy.

The document maps seven archetypal strategies into three families.

The coexistence family includes dominance, in which the United States leads AI development and suppresses rivals; co-development, in which the United States leads a consortium that includes China under shared governance, pooled compute, verification and monitoring; and preparedness, which relies on informal coordination, voluntary commitments, phased deployment, independent evaluation, incident reporting and domestic resilience.

The denial family is more restrictive.

A moratorium would try to verify a worldwide stop to AI work above defined danger lines.

Deterrence would pause U.S. frontier work past its own thresholds while using national power to block rival programs.

The continuity-of-society option is a last-resort survival model, built around dispersed and self-sustaining communities designed to retain human agency if prevention and coexistence fail.

Acceleration sits in its own family.

It assumes that constraining development is more dangerous than AI development itself, so markets, competition and rapid iteration become the route to safety.

That option is included in the framework rather than dismissed, but RAND’s broader point is that a government choosing acceleration still needs to know what it is not building: restraint tools, verification systems, institutional capacity and crisis plans.

The decision tests show why the strategy is framed as optionality.

Policymakers would have to judge how close AI danger is, whether human-AI coexistence looks feasible, whether restraint can be coordinated, whether a decisive strategic advantage is possible and whether domestic or foreign AI programs can be suppressed through controls.

Different answers point toward different families of action.

Those tests also make the policy problem operational rather than abstract.

If danger appears close, defensive capacity and resilience become more important.

If coexistence appears workable, development and deployment strategies become easier to justify.

If restraint is feasible, cooperative approaches are available; if it is not, coercive enforcement becomes the only restraint path left.

If no single actor can hold a decisive AI lead, a dominance strategy loses much of its logic.

Import AI’s own reading is that current U.S. posture resembles acceleration: more resources go into making the AI “car” faster than into seatbelts, headlights or brakes.

The policy consequence is concrete.

Keeping options open requires public spending before the moment of crisis, because verification systems, safety research, compute visibility, institutional expertise and societal response plans cannot be improvised after frontier systems have already changed the strategic environment.