Responsible AI Statement

01The commitment

AI assists, humans author. Our AI drafts, suggests and searches. It never signs, never overrides a professional, and never issues anything client-facing on its own. Every AI output in our products shows its source and requires a named person's review and acceptance.

02The models we use

AI features are powered by Anthropic (Claude) and OpenAI models via their commercial APIs. Under those commercial terms, customer data is not used to train the providers' models. Provider-side API retention is short-lived (typically up to 7-30 days for abuse monitoring) and is listed in our Data Retention Schedule.

03Guardrails in the product

• Human review is a workflow step, not an option: unreviewed output cannot be issued as final
• AI suggestions cite the source they were derived from (your data, standards structures, your rate history)
• AI features degrade safely: if a response is truncated, refused or malformed, the product shows nothing rather than something wrong
• Capabilities are labelled honestly: Live, In development, or Roadmap, on the product pages and in the app

04Limitations, stated plainly

AI can be wrong, incomplete, or plausible-but-mistaken. That is exactly why the human sign-off step exists and why the Professional Use & Reliance Statement is part of our terms. Treat every AI output as a draft for professional judgement.

05Contestability and change

If an AI feature behaves in a way you think is wrong or unfair, tell us: hello@obedience.global. We log, investigate and answer. When we change the models or providers behind a feature, we record it in the policy changelog. This statement will be updated as the ICO's statutory Code of Practice on AI and automated decision-making is finalised.