What it means
- A person affected by an AI-driven outcome can easily ask for a human to look at it again, and can submit facts the AI may have missed.
- The human reviewer can change the outcome, and the person is told the result.
- Customers can switch from an AI channel to a human at any point in regulated journeys.
- ISO 27001 reuse: None. New for AI, not covered by ISO 27001. SOC 2 P5.2 (correcting personal information) is a partial analogue.
Why it matters
- RBI FREE-AI expects customers to always be offered a switch to a human when dealing with AI.
- Wrong automated decisions (blocked accounts, rejected claims, failed KYC) cause real harm and complaints that escalate to the RBI Ombudsman.
- Laws abroad (for example EU GDPR Article 22 and the EU AI Act) give individuals similar rights; customers of Indian SaaS exporters will ask for it.
Checklist
- Identify AI outcomes that significantly affect individuals (credit, insurance, employment, account restrictions, KYC).
- Add a clear "request human review" link or option in the notice of outcome and in the AI chat or voice flow.
- Define who reviews, required skill, and a target turnaround (for example 7 working days, faster for account blocks).
- Give reviewers the full case: inputs, model output, explanation and the person's new information (SC-AI-024 Explain AI-assisted decisions).
- Make sure the reviewer is not just re-running the same model.
- Tell the person the outcome and the reasons, and point to the grievance route if they remain unhappy (SC-AI-028 Customer grievance redressal for AI).
- Track review volumes, reversal rates and turnaround; feed reversal patterns into model improvement (SC-AI-020 Performance monitoring and drift detection).
Evidence examples
- Screenshots of the "request human review" option in notices and chat flows.
- Human review SOP with roles and turnaround targets.
- Case log showing requests, reviewer, outcome, reasons and dates.
- Monthly report of reversal rate and turnaround versus target.
- Test record showing the AI-to-human switch works in each channel.
Common pitfalls
- Review request hidden behind several menus or only available by post.
- Reviewer sees only the AI score and confirms it.
- No feedback loop, so the same error repeats for thousands of customers.
- Human handoff available only during office hours with no notice of when to expect contact.
Satisfies
- BindingEU AI Act
- BindingNYC Local Law 144 (AEDT)
- BindingKorea AI Basic Act
- VoluntaryRBI FREE-AI
Standards crosswalk
Clause references only. Check the standard itself for the exact wording.
ISO/IEC 42001
- A.8.2 system documentation and information for users
- A.8.3 external reporting
- A.9.2 processes for responsible use
ISO/IEC 27001
No direct equivalent
SOC 2
- CC2.3
- P5.2
Key takeaways
- Make asking for a human review easy, at the point the outcome is shown.
- The reviewer must have the full case and the power to change the result.
- Use reversals as a signal to fix the model.