AGI as National Security

The argument that AGI and superintelligence are fundamentally national security challenges rather than consumer technology products. Whoever leads in superintelligence gains decisive military advantage, comparable to (or exceeding) the advantage nuclear weapons confer.

The Core Argument

As laid out in situational-awareness by leopold-aschenbrenner:

Decisive military advantage. Superintelligence applied to military R&D could compress a century of technological progress into years. The Gulf War demonstrated what a 20-30 year tech lead means: total dominance despite numerical parity. A superintelligence lead would be far more decisive -- potentially enough to neutralize nuclear deterrents through advanced sensor networks, autonomous drone swarms, and impenetrable missile defense.

Economic explosion. Automated labor and scientific research could push growth rates into double digits annually. The GDP of whoever leads would rapidly dwarf everyone else's.

Authoritarian peril. A dictator with superintelligence could permanently lock in power through AI-controlled robotic police, mass surveillance, and perfect loyalty enforcement. Unlike past dictatorships, superintelligence could eliminate all historical threats to authoritarian rule (coups, rebellions, succession crises), enabling permanent totalitarian control.

China Can Be Competitive

Despite chip export controls and US lab dominance, China has a credible path:

1. Chips: SMIC can produce 7nm chips (~2-3x worse perf/$ than Nvidia, but functional)
2. Infrastructure: China has built as much new electricity capacity in the last decade as the entire current US capacity. Building 100GW clusters is easier for China.
3. Algorithms: On the current course, Chinese espionage can trivially steal the key algorithmic breakthroughs from poorly-secured US labs, potentially matching the US algorithmic edge.

The Security Crisis

AI lab security is catastrophically inadequate. Two assets need protection:

• Model weights: The finished AI system (a large file that can be copied). Stealing the automated-AI-researcher weights on the cusp of an intelligence explosion would be the ultimate prize.
• Algorithmic secrets: The breakthroughs making AGI possible. More urgent to secure because they're being developed now. Worth 10-100x compute advantage.

Current security is "random startup" level. Labs admit to being at the lowest tier of their own security frameworks. No state-actor-proof defenses exist at any US AI lab.

The Project

Aschenbrenner predicts a government-led AGI project ("The Project") by 2027-28, analogous to the Manhattan Project. Not literal nationalization but a Boeing/Lockheed-style defense relationship: labs "voluntarily" merge into the national effort, Congress appropriates trillions, international democratic coalition forms (modeled on the Quebec Agreement), nonproliferation regime established (modeled on NPT/IAEA).

Implications

• AGI clusters must be built in the US, not Middle Eastern autocracies
• AI lab security must reach state-actor-proof levels, requiring government help
• Democratic coalition needed to maintain lead and develop nonproliferation norms
• A healthy lead (years, not months) provides the margin needed for both safety and stability
• The civilian applications of superintelligence come after the initial national security period, similar to nuclear energy following nuclear weapons

Sources

• situational-awareness - Parts IIIb, IIId, IV