The US is betting on a "Freedom of Action" strategy to handle the road to superintelligence

RAND recently laid out a massive framework for how the US can maintain a geopolitical edge while navigating the transition to superintelligence. The core logic is that since the actual trajectory of AI takeoff is a total mystery, the only logical move is to avoid locking into one single path. Instead, they're proposing a "Freedom of Action" strategy designed to ensure human agency and national survival regardless of how the tech evolves.

What makes up the Freedom of Action Strategy?

To keep these options open, RAND identifies four critical pillars that need simultaneous investment:

• Human-AI Ecosystem: This involves funding AI safety, creating tools that prevent humans from losing agency, and preparing the general public for the economic and social disruptions AI will trigger.
• AI-Security Architecture: The goal here is visibility. The US needs tools to monitor the frontier of AI systems globally, specifically gaining insight into underlying compute and creating ways to verify international agreements.
• National Security Overhaul: Legacy security institutions need a complete update to actually utilize AI capabilities and survive the changes the tech brings to warfare and intelligence.
• Response Capacity: This is about resilience—improving AI literacy for decision-makers and building "break-glass" plans for when things actually go wrong.

Seven archetypal strategies across three families

The paper categorizes potential paths into three distinct families based on how the US interacts with other global powers and the AI itself:

The Coexistence Family

• Dominance: Leading the charge in development while actively suppressing competitors.
• Co-Development: Forming a consortium (including China) to pool compute and share governance for safe superintelligence.
• Preparedness: Using informal coordination, voluntary commitments, and incident reporting to manage risks.

• Moratorium: Pushing for a verifiable global stop on AI development once specific thresholds are hit.
• Deterrence: Halting domestic frontier development above certain limits while using national power to disrupt foreign programs.
• Continuity of Society: The "worst-case" plan. If coexistence fails, the US would create geographically distributed, biologically self-sustaining settlements hardened against AI threats.

• Rapid Iteration: This assumes that slowing down is riskier than speeding up. It relies on market competition and fast cycles to treat safety as a byproduct of creating useful AI.

The five main uncertainties

The strategy acknowledges that the "right" choice depends on variables we can't currently quantify. The most critical is Danger Proximity: if superintelligence becomes dangerous tomorrow, we need defenses and resiliency now. If the danger is far off—or if we are in more danger by not progressing—then acceleration becomes the priority. Another massive unknown is Coexistence Feasibility, or whether humans and superintelligent systems can even find a stable equilibrium.