Responsible AI

AI can help read evidence. It does not decide what is true about a person.

Arrota uses AI to assist document extraction and classification. A person reviews the output, and AI does not determine whether a record is valid.

How trust actually works

Where AI helps

Reading, not judging.

  • Extracting text from documents a holder uploads, where supported
  • Identifying likely fields such as dates, titles, and organisations
  • Structuring that information for a person to review
  • Suggesting a classification for a person to review

Where it stops

Four decisions AI does not make.

  • ExcludedDecide whether a credential is genuine
  • ExcludedTurn a holder upload into Organisation-confirmed proof
  • ExcludedScore a person’s trustworthiness, employability, talent, or credibility
  • ExcludedInfer facts about a person beyond the submitted document

People in the loop

High-confidence extractions are stored automatically.

High-confidence extractions are stored in the holder's vault automatically. Fields the model is unsure about are marked for review, and a holder can correct anything in their vault at any time — but review is a path for uncertain fields, not a gate every record waits behind.

Boundaries in the Trust Center

Data, models and feedback

How Arrota handles data, models, and feedback.

Data handling

AI processes a holder’s document only for the purpose for which it was submitted. Personal data is not used to train models. The Privacy Notice describes the approved handling.

Claim matching

Formal verification matches structured fields deterministically. It is not an AI judgment. AI may assist extraction or normalisation; validity requires traceable source evidence.

Model changes

When AI behaviour changes materially, this page and the Trust Center will be updated.

Feedback

If a document was read incorrectly or a boundary is unclear, tell us . AI-related corrections follow the same traceable process as record corrections.