Responsible AI
Principles are easy to publish. We write ours as constraints, because that is how they get built.
Position
Stated as constraints
Principles are easy to publish. These are written as constraints because that is how they are implemented.
| Constraint | What it means in practice |
|---|---|
| AI assists, humans decide | No clinical, regulatory or programme decision is made by a model without a named human accepting it. |
| Grounded, not recalled | Agents reason over retrieved, approved reference material rather than free recall. |
| Explainable at the point of use | Every AI output carries a visible reason, not a score alone. |
| Reversible | Any AI-influenced change can be undone by the human who owns it. |
| Auditable | Inputs, outputs and rationale are written to an immutable log. |
| Removable | The platform must work with the AI layer disabled. If it cannot, the AI is load-bearing in a way we do not accept. |
| Validated before use | Performance is established on representative data before deployment, and monitored for drift afterwards. |
| Privacy by design | Data minimisation, access control and consent are architectural, not procedural. |
On compliance language
CAD CARE describes its architecture as designed to support the requirements of frameworks such as India's Digital Personal Data Protection Act, and — where applicable to an engagement — HIPAA, GDPR, ISO 27001 or SOC 2. Design alignment is not certification, and this website does not claim certification.
Ask us the hard question.
Bring the failure mode you are worried about. That is the right place to start an AI conversation.