Absent a single federal AI statute, US obligations arrive state by state. The common thread across the serious ones is documentation: developers and deployers must be able to show their work on consequential automated decisions.
The first comprehensive US state AI law. It imposes a duty of reasonable care on developers and deployers of high-risk AI systems — those making, or substantially factoring into, consequential decisions (employment, lending, housing, healthcare, insurance, education, legal services). Core duties include impact documentation, disclosure to affected consumers, and the ability to explain the role AI played in an adverse decision.
Explaining the role AI played in a specific adverse decision is exactly what a per-decision causal record enables: not a general model description, but the specific configuration that governed that output.
Beyond Colorado, a growing set of state measures converge on the same evidentiary need:
Each speaks a different legal dialect, but each ultimately asks the operator to produce durable, per-decision evidence of how an automated decision was reached.