Clinical decisions
Diagnosis, prescribing, medication changes, clinical interpretation, and urgent safety response stay with your licensed team.
How Rely works / Safety and governance
Your licensed team retains every clinical decision.
Rely routes novel cases to review and releases workflow changes only after testing, approval, and monitoring.
A production change has to earn its way through.
New behavior moves through review, testing, approval, deployment, and monitoring.
- 01Change proposed
- 02Risk and scope reviewed
- 03Cases tested
- 04Owner approves
- 05Controlled release
- 06Monitor or roll back
Novel clinical cases go to human review. The model does not expand its own authority.
A production change has to earn its way through. New behavior moves through review, testing, approval, deployment, and monitoring. The diagram shows Change proposed, Risk and scope reviewed, Cases tested, Owner approves, Controlled release, Monitor or roll back. Novel clinical cases go to human review. The model does not expand its own authority.
Boundaries
What AI does not own
Novel exceptions
An unusual case enters review; it does not create a new rule.
Fallback and accountability
If a dependency fails, the configured fallback runs and the owner stays visible.
Production changes move through control
- 01
Evidence and proposal
A reviewed trace, exception, or operating result supports a written change.
- 02
Regression test
The proposal is tested against routine cases, edge cases, stops, and fallbacks.
- 03
Named approval and controlled release
The accountable clinical or operational owner approves the change before a bounded release.
- 04
Post-release monitoring
The changed path is watched for regressions and rolled back when it does not behave as approved.
Clearance
Program clearance is not company security posture
Company security posture does not clear a workflow. Each program separately approves its data, communications, clinical boundary, testing, fallbacks, and monitoring.
- Data: minimum necessary fields, access, retention, and reconciliation.
- Communication: AI disclosure, recording, consent, cadence, and language.
- Clinical and operational: actions, stops, owners, fallbacks, testing, and monitoring.
Bring reviewers in early.
Agree the boundary, escalation path, and clearance list before setting a launch date.