10 to the 23 AI logo

Real-World AI Governance Center

Responsible AI for High-Stakes Human Systems

AI is already making and shaping decisions inside child welfare, public benefits, casework, and other systems where the people affected have the least power to contest an error. This Center treats that as a governance problem: not "is the model good?" but "does this whole sociotechnical system — model, people, records, pressures — catch its errors faster than it spreads them?"

Five sections, one discipline: concepts in the Field Guide, documented histories in the Domain Atlas, levers in the Practice Library, stress testing in the PAN Lab, and every empirical claim ledgered in the Evidence Registry.

Three ways in

New to the topic

Field Guidea case filethe PAN Lab

Get the concepts, see one documented history, then watch governance choices move a stylized system.

Leading a deployment

Domain AtlasPractice LibraryWork With 1023AI

Find your domain and its failure modes, shortlist the levers, then talk about your system's actual shape.

Checking our homework

Evidence RegistryGovernance Watch

Start from the sources and work backwards. Every empirical claim is ledgered; pending citations are tracked openly.

The evidence discipline, in short

Every empirical claim on this site is a ledger entry mapped to sources synced from the PAN reference library — never added by hand. Conceptual framing is labeled as framing; scenario results are labeled as illustrative PAN-run results, direction and shape only; and statements still awaiting a source carry a visible "citation pending" badge rather than quiet confidence.

The full registry — including what's pending — is public at Evidence Registry.