PubLedge

Beginning January 1, 2027, a developer of covered automated decision-making technology must disclose to deployers the technology's intended uses, the categories of data used to train it, its known limitations, and instructions for meaningful human review. The duty runs developer-to-deployer, not developer-to-consumer.

What Counts

• Documentation naming the intended uses the developer supports for the ADMT
• Description of training-data categories at a level that lets a deployer assess fit
• Statement of known limitations, including populations or inputs where performance is unreliable
• Instructions sufficient for a deployer to conduct meaningful human review of an output

What Does Not Count

• A marketing datasheet with capability claims but no limitations
• Documentation available only to consumers rather than to the deploying entity
• A generic model card that omits intended uses or human-review instructions
• Deferring the disclosure until after the deployer has placed the ADMT in service

Implementing Legal Instruments