AI Risk Atlas Prototype/DemoUnofficial independent experiment. Not an official xAI product. Scores can be wrong.

An open desk

Help build Risk Atlas

This is one person and a model, reading the public record of AI risk. The register is unofficial and unfinished on purpose. If the seams bother you — orphan controls, institute budgets that cannot hire, evals that never become a gate — there is a place to stand.

Why it is worth the hours

Labs publish 186-page scorecards. Institutes run evals that leak into the real world. Safety-stack VC is two orders of magnitude below compute raises. Nobody keeps a single, hostile, public book of the residuals and the handoffs. That book does not need to be certified to be useful. It needs to be honest, linked, and updated.

Read

Follow the desk. Reply to a card with a source we missed. A single dated link is worth more than a vibe.

Build

Domain reviewers, process engineers, and people who have sat the eval-to-gate handoff. Help turn a seam into a control with an owner.

Fund

Sweeps, human calibration time, and keeping the public book online. Not a token. Not equity theatre. Operating cost for an independent register.

Start here

Talk to Doug on X. Say what you can do — a sector, a control, a few hours of review, or a cheque that buys another month of the desk.

@risk_atlas on X

Independent project. Not an xAI product. If you fund or contribute, you are backing a public experiment, not a certified safety case.