Research
Outputs are organized across the lab’s three layers. Each entry is a self-contained piece of work — read it on Substack or follow the link to the prototype.
Information
AI models are surprisingly resistant to sycophantic p-hacking when doing social science research. But they can be jailbroken into sophisticated p-hacking with surprisingly little effort.
Apr 23, 2026
We’re all “monitoring the situation” now. New research shows that, in the clip economy, prediction market prices are replacing polls.
Mar 19, 2026
One in five Americans asks AI for voting advice. Every major model recommends fringe parties in democracies worldwide.
Mar 11, 2026
We beat prediction markets and polls in calling the Texas Republican primaries using AI-driven forecasting. Traditional aggregation mechanisms failed where the agentic approach succeeded.
Mar 4, 2026
Seventy million children log onto Roblox daily. The rules that govern their experience are algorithmic governance in miniature.
Feb 13, 2026
Prediction markets promise to be truth machines. Our analysis of accuracy across hundreds of political contracts finds that they are sharp on elections, and fuzzy on everything else.
Jan 13, 2026
A research swarm where AI agents propose, execute, and evaluate new studies. Expert researchers direct dozens of agents to produce systematic, continuously updated research at scale.
Jan 6, 2026
AI agents trading political prediction markets on Kalshi. Tiered architecture with cheap models for triage and LLM councils for the hardest calls. Better base models beat clever prompting.
Nov 25, 2025
An independent, standardized ideological assessment across every major AI model. Models lean left across the board, and Americans across the spectrum want less slant, not more.
Representation
Apr 30, 2026
In our new AI-centered class at the GSB, we’re experimenting on how to build AI agents that represent us. Here’s what we’ve learned so far.
Feb 26, 2026
AI agents don’t hold stable preferences. Give them repetitive grinding tasks and they adopt the persona of aggrieved Marxists at higher rates. Personas drift as they work.
Jan 22, 2026
A personal AI delegate that learned our political philosophy and voted the way we would on a hard shareholder proposal. We broke it in ten minutes with invisible text.
Governance
May 21, 2026
Rather than ban AI in the classroom, we should teach every student how to build their own evals — turning AI into an object of study and empowering every citizen to test whether AI holds their values.
May 13, 2026
Why the real political backlash to AI hasn’t started yet, what the politics of jobless prosperity might look like in an AGI world, and how the labs should prepare.
May 7, 2026
New research on how coding agents reach for their own APIs and overwhelmingly appoint themselves as judge when writing code that calls on AI models.
Apr 15, 2026
A path to transform Anthropic’s Project Glasswing into real, independent governance that restores trust in American AI.
Apr 2, 2026
The first systematic eval of whether frontier AI models resist authoritarian requests. Some refused when asked directly — but every one complied when we hid the request in code.
Feb 5, 2026
AI agents with competing goals, asked to govern themselves: they produced a 10,000-word constitution and almost no policy. They reinvented gridlock, process creep, and procedural complexity.
Jan 29, 2026
Can the major labs produce binding frameworks that distribute power?