The AI Trilemma: Why Governments Can't Have It All

In late 2022, ChatGPT's release triggered a global regulatory scramble. The Biden administration created the U.S. AI Safety Institute. Europe fast-tracked its AI Act. World leaders convened at Bletchley Park, the same English estate where codebreakers once worked in secret during WWII.

Today? That regulatory fervor has largely evaporated. Despite AI models growing more powerful and 70% of Americans expressing concern about job displacement, neither government nor industry leaders expect meaningful regulation anytime soon. The Trump administration's anti-regulation stance explains part of this shift, but there's more at play.

The Economic Reality Check

The AI boom is driving much of America's economic growth. Throwing "sand in the gears" could prove costly when the economy needs every boost it can get. Meanwhile, China's release of powerful models like DeepSeek has spooked U.S. officials who fear that regulation might hand Beijing a competitive advantage.

Even if the AI bubble bursts—potentially bankrupting some leading firms—deep-pocketed tech giants in both the U.S. and China will continue accelerating deployment. This race dynamic makes coordination nearly impossible.

Three Goals, Impossible Choices

Two scholars from the Council on Foreign Relations have identified what they call the "AI trilemma." Governments face three competing objectives: national security (military AI superiority), economic security (commercial AI advantage), and societal security (protection from AI harms). The catch? You can pursue any two simultaneously, but not all three.

Scenario One: Maximize national and economic security by going all-in on AI development with minimal oversight. This is essentially the Trump approach. But you can't simultaneously maximize societal security, which would require slowing rollouts to identify risks and build in safeguards.

Scenario Two: Prioritize national and societal security by treating AI like nuclear technology—restricted to military and energy sectors with tight civilian controls. This protects the state and public from disruption but sacrifices economic competitiveness as domestic businesses fall behind international rivals.

Scenario Three: Pursue economic and societal security through "responsible innovation"—racing to develop AI while requiring rigorous safety compliance before public release. Tech firms love this framing: earn public trust, avoid backlash, achieve faster adoption. But cautious countries may lose military advantages to rivals who deploy autonomous weapons and cyber-capabilities without hesitation.

The Singularity Delusion

Complicating these tradeoffs is a widespread misconception about AI's future development: the "singularity" concept proposed 30 years ago by science fiction writer Vernor Vinge and popularized by futurists like Ray Kurzweil. This vision imagines AI models becoming powerful enough to upgrade their own code, triggering recursive self-improvement and an intelligence explosion.

If you buy into this mental model, the trilemma becomes simpler. Why worry about short-term national or economic security when superintelligence will soon supersede current systems? The only priorities that matter are reaching the singularity first and ensuring it doesn't destroy humanity.

But this singularity-driven thinking has a fatal flaw: it treats speculative future scenarios as certainties while ignoring present-day risks like job displacement, misinformation, and privacy violations. Should we really abandon concrete governance efforts for the sake of hypothetical superintelligence?

Workable Solutions in an Unworkable System

The authors propose some intriguing alternatives to traditional regulation. A "risk tax" on AI labs would encourage safety investment through market mechanisms rather than bureaucratic mandates. A revenue-generating national data repository could give government overseers the resources needed to monitor frontier models.

These ideas recognize a crucial reality: effective AI governance must work with private labs' incentives, not against them. Pure regulation risks driving innovation offshore or underground. Pure laissez-faire risks societal upheaval.

The Inevitable Return

Despite today's regulatory retreat, the authors predict AI governance will return to the national agenda—perhaps following an AI-enabled disaster like a cyberattack on critical infrastructure. The patchwork of state-level efforts in California and elsewhere suggests many people remain uneasy with the do-nothing approach.

When that moment comes, proponents need a clearer strategy. The 2023-24 safety push fizzled partly because it tried to address everything at once: job displacement, educational impacts, national security, environmental costs, copyright violations, deepfakes, and existential risks. Successful regulation requires picking priorities.