What it means
- Tell people plainly, at the start, when they are talking to or being served by an AI system (chatbot, voice bot, AI email agent, AI-drafted reply).
- Tell them about known limits: AI output can be wrong or unreliable, and they should verify important information.
- Where a law or regulator requires it, offer an easy route to a human.
- ISO 27001 reuse: None. New for AI, not covered by ISO 27001. Privacy notice processes (SOC 2 P1.1) give a starting template.
Why it matters
- RBI FREE-AI asks regulated entities to tell customers when they are dealing with AI and to always offer a switch to a human.
- The MeitY March 2024 advisory asks platforms to label under-tested or unreliable models and to use a consent pop-up about possible unreliability.
- Hidden AI erodes trust fast; it also undermines any later defence that the customer knew what they were dealing with.
Checklist
- List every customer-facing and employee-facing touchpoint where AI talks to people (from SC-AI-003 AI system inventory).
- Write standard disclosure text for each channel: chat banner, voice greeting, email footer, in-app tooltip. Keep it in English plus the main regional languages you serve.
- Add a reliability notice for generative features ("AI can make mistakes; check important details").
- For beta or under-tested models, add a visible "experimental" or "may be unreliable" label and a one-time acknowledgement pop-up.
- Provide a visible "talk to a human" option in BFSI and other regulated journeys; test that it works (SC-AI-027 Right to contest and human review for individuals).
- Never let an AI agent claim to be a human, including in outbound voice calls.
- Add a release checklist item so new AI features cannot ship without the disclosure.
- Review wording yearly and after complaints.
Evidence examples
- Screenshots of the chat banner, voice script and pop-up in production.
- Approved disclosure text library with version history.
- Product release checklist showing the disclosure check passed.
- Test record showing "talk to a human" hands off correctly.
- Consent or acknowledgement logs for the unreliability pop-up.
Common pitfalls
- Disclosure buried in terms of service instead of shown at the point of interaction.
- Voice bots with human-sounding names and no AI announcement.
- Human handoff exists on paper but loops back to the bot.
- English-only disclosures for customers served in Hindi or regional languages.
Satisfies
- VoluntaryISO/IEC 42001
- BindingEU AI Act
- BindingNYC Local Law 144 (AEDT)
- BindingTexas TRAIGA
- BindingKorea AI Basic Act
- SupervisoryMeitY AI and Deepfake Advisories
- VoluntaryRBI FREE-AI
Also relevant to
- ICMR AI in Healthcare Guidelinesguidance mapping (patient choice; no obligation IDs)
Standards crosswalk
Clause references only. Check the standard itself for the exact wording.
ISO/IEC 42001
- 7.4 communication
- A.8.2 system documentation and information for users
- A.8.5 information for interested parties
ISO/IEC 27001
No direct equivalent
SOC 2
- CC2.3
- P1.1
Key takeaways
- Say "this is AI" at the start of the interaction, in the user's language.
- Add a plain reliability warning for generative features and label beta models.
- In regulated journeys, a working human handoff is part of the disclosure.