Cooperation doesn't run on intelligence alone

Stanford’s 2026 AI Index found that frontier models gained 30 percentage points in a single year on Humanity’s Last Exam — a benchmark built to be hard for AI and favorable to human experts.

Alignment needs to keep up with capabilities if we’re to stay in control, prevent bad actors from turning the same technology that could cure cancer into a bioweapon, and make sure AI serves society as a whole, not just a subset of it. 

The tension between pacing and racing has no easy resolution when the same capabilities are both offensive and defensive. The tools the industry has built are serious. The ceiling they keep hitting is a structural property of every approach. 

Every one of them works on the same substrate: intelligence. Alignment work focuses on constraining it. 

Meanwhile, the capability that cooperation actually runs on has received a fraction of the attention. What if that’s the missing piece?

This series follows that thread. The last piece is a proposal for making alignment rational – specific enough to be tested and wrong, with an MVP narrow enough to test the direction.