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

The current public assistants use Anthropic through its commercial API for low-sensitivity product and policy questions. Quantify contains optional Anthropic and OpenAI integration code, but OpenAI private processing and confidential hosted-model processing are not approved current paths. A credential, dependency or feature flag is not approval.

By default, commercial API providers do not use commercial API inputs or outputs to train their models. Standard API content or abuse-monitoring retention can be up to 30 days. Provider legal and safety exceptions may retain limited flagged records for longer, and endpoint-specific application state can have a different lifecycle. Current provider, approval and endpoint controls are recorded in our Data Retention Schedule and Service Provider and Subprocessor List.

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.