AI & personal dataUpdated 22 November 20269 min read

AI Products for Children: DPDP Tracking and Profiling Limits

What section 9 and Rules 10 and 12 mean for AI tutors, companion chatbots, kids' apps and any AI product that may be used by someone under 18 in India.

By Suhas Bhairav · Reviewed against the DPDP Rules, 2025 (G.S.R. 846(E), 13 Nov 2025)

Age of a child
Under 18 years
Parental consent rule
Rule 10 (verifiable consent)
Prohibited
Tracking, behavioural monitoring, targeted ads (s. 9(3))
Exemptions
Rule 12 and Fourth Schedule Parts A and B
Maximum penalty
Up to ₹200 crore
Applies from
13 May 2027

Quick answer

Under section 9 of the DPDP Act, an AI product processing data of anyone under 18 needs verifiable parental consent under Rule 10, must not process data in ways likely to harm a child's well-being, and must not track, behaviourally monitor or target ads at children. Fourth Schedule exemptions are narrow. These duties apply from 13 May 2027, with penalties up to ₹200 crore.

  • A child under DPDP is anyone under 18, so teen users of AI apps are children.
  • Rule 10 requires due diligence that the person giving consent is an identifiable adult parent, using held data, provided details or a virtual token.
  • Section 9(3) bans tracking, behavioural monitoring and targeted advertising directed at children; adaptive AI memory and engagement profiling need careful design.
  • Section 9(2)'s ban on detrimental processing has no Fourth Schedule exemption.
  • Breach of children's obligations carries a penalty of up to ₹200 crore per instance.

Who this is for: Founders of kids' and teen AI apps, AI tutor and EdTech product teams, Companion and chatbot product teams, Trust and safety teams, Data Protection Officers

The three section 9 duties for AI products

Section 9 of the DPDP Act sets three duties for any Data Fiduciary processing a child's personal data. A child is an individual who has not completed eighteen years.

  1. Verifiable parental consent (s. 9(1)). Before processing a child's personal data, obtain verifiable consent of the parent (or lawful guardian). Rule 10 sets out how.
  2. No detrimental processing (s. 9(2)). Do not process personal data in a manner likely to cause any detrimental effect on the well-being of a child.
  3. No tracking, behavioural monitoring or targeted advertising (s. 9(3)). Do not undertake tracking or behavioural monitoring of children, or targeted advertising directed at children.

Rule 12 and the Fourth Schedule switch off sections 9(1) and 9(3) for specific classes of fiduciary and purposes, under conditions. Section 9(2) is never switched off.

For AI products, section 9(3) is the hardest constraint. Many AI features are, technically, behavioural monitoring: long-term memory of a user's conversations, engagement-optimised recommendations, sentiment tracking, learning-style profiling. The question for each feature is whether it monitors a child's behaviour over time, and if so, whether a Fourth Schedule condition genuinely covers it.

Section 9(5) lets the Government lower the age threshold for fiduciaries that process in a verifiably safe manner; no such notification is assumed here.

Rule 10(1) requires technical and organisational measures to ensure verifiable parental consent before processing any personal data of a child, and due diligence to check that the person identifying herself as the parent is an adult who is identifiable, if required under law. The check is made by reference to:

  • (a) reliable details of identity and age already available with the fiduciary; or
  • (b) details of identity and age voluntarily provided by the individual, or through a virtual token mapped to such details issued by an authorised entity — which includes details or tokens made available and verified by a Digital Locker service provider.

The Rule's four illustrations cover two entry paths — the child declares a parent, or the parent opens the account — and two parent states — already a registered user with reliable details held, or not registered, in which case the fiduciary checks identity and age details issued by an entity entrusted by law or Government, or a virtual token, which the parent may provide through DigiLocker.

Engineering pattern:

  1. Age gate at sign-up; under-18 path pauses data collection beyond what is needed to contact the parent.
  2. Parent flow: log in (if existing verified adult) or verify via token.
  3. Store consent record: parent ID reference, child account, notice version, timestamp, verification method — not the parent's ID document.
  4. Unlock the child account only after consent.

Implementation details are in age verification and DigiLocker tokens.

Reviewing AI features against section 9(3)

Run every AI feature through a simple test: does it build or use a persistent picture of this child's behaviour?

AI featureRisk under s. 9(3)Safer design
Long-term chat memoryBehavioural monitoringSession-only memory, or parent-visible, purpose-limited memory
Engagement-ranked feedTracking and monitoringContextual or curated content
Personalised adsTargeted advertising (prohibited)No ads, or contextual non-targeted ads
Emotion or sentiment scoringBehavioural monitoringDo not deploy for children
Adaptive learning difficultyMonitoring, unless an exempt educational institutionWithin-session adaptation; institution-led programmes
Safety classifiers on messagesPossibly exempt (Part B item 5)Restrict to blocking harmful content
Location featuresTrackingReal-time safety location only (Part B item 4)

Inference-time adaptation within a single session — adjusting explanations based on the current question — is much easier to defend than a cross-session profile.

Also look at telemetry: product analytics SDKs that build user-level behavioural funnels for under-18 accounts can breach section 9(3) even if the AI feature itself is clean.

What the Fourth Schedule allows

Rule 12 disapplies sections 9(1) and 9(3) in two ways.

Part A — classes of fiduciary, each with a condition:

  • Clinical establishments, mental health establishments and healthcare professionals — restricted to health services to the extent necessary to protect the child's health.
  • Allied healthcare professionals — restricted to supporting a recommended treatment and referral plan.
  • Educational institutions — tracking and behavioural monitoring restricted to educational activities or safety of enrolled children.
  • Crèche or child day care carers — tracking and monitoring for safety.
  • Transport providers engaged by those institutions — location tracking during travel to and from them.

Part B — purposes, each restricted to what is necessary:

  1. Exercising legal powers or duties in the child's interest.
  2. Subsidies, benefits, services, certificates, licences or permits under section 7(b).
  3. Creating a user account used only for communicating by email.
  4. Determining real-time location for the child's safety, protection or security.
  5. Ensuring information, services or advertisements likely to harm the child are not accessible to her.
  6. Confirming a user is not a child, and observing Rule 10 due diligence.

A consumer AI app is usually not in Part A. Part B items 5 and 6 are the ones most AI products use: safety filtering and age assurance. Full detail in children's data exemptions.

Designing against detrimental effect

Section 9(2) prohibits processing likely to cause any detrimental effect on a child's well-being, with no exemption. The Act does not define "detrimental effect", so build a documented safety case.

For generative AI products used by children, that includes:

  • Content safety. Classifiers on inputs and outputs for self-harm, sexual content, violence and dangerous instructions; refuse and redirect to help resources.
  • Companion dynamics. Avoid designs that encourage dependency, secrecy from parents or excessive use; cap session lengths and avoid manipulative engagement prompts.
  • Data minimisation in conversations. Discourage children from sharing addresses, school names or photos; mask what they share before logging.
  • Human escalation. Route serious safety signals to trained reviewers under a documented protocol.
  • Logging and review. Keep masked safety logs to investigate incidents; Rule 8(3) requires processing logs for at least a year.

The MeitY India AI Governance Guidelines list "people first" and "safety, resilience and sustainability" among their sutras; they are principle-based rather than binding, but they signal how regulators will view child-facing AI.

What to do before 13 May 2027

  • Decide your audience honestly. If teens use your general AI product, you process children's data. "18+ only" in terms of service without age assurance is weak.
  • Ship age assurance. Use Part B item 6 to confirm users are not children, with minimal data.
  • Build the parental consent flow under Rule 10, with DigiLocker or other token-based verification for unregistered parents.
  • Split the feature set. Under-18 accounts get session-only memory, no behavioural ranking, no targeted ads.
  • Audit SDKs. Remove ad and analytics SDKs that profile child accounts.
  • Write the safety case for section 9(2).
  • Update notices under Rule 3, written for parents.
  • Mask and expire logs. Mask children's prompts before they reach traces, keep processing logs for the Rule 8(3) year, then erase.
  • Test with families. Run the parental consent flow with real parents before launch and fix drop-off points.

EdTech businesses should also read children's data in EdTech AI, and platforms with recommendation engines AI personalisation under DPDP.

Further reading on suhasbhairav.com

Before 13 May 2027

Compliance checklist

  1. Implement an age gate with minimal data

    Confirm whether a user is a child before full processing, using only what is necessary.

    Fourth Schedule Part B item 6

  2. Build verifiable parental consent

    Verify the parent is an identifiable adult using held data, provided details or a virtual token.

    Rule 10(1)

  3. Support DigiLocker-based verification

    Accept identity and age tokens from a Digital Locker service provider for unregistered parents.

    Rule 10(2)(b)

  4. Disable cross-session behavioural profiling for children

    Use session-only memory and contextual content for under-18 accounts.

    DPDP Act s. 9(3)

  5. Remove targeted advertising for children

    Turn off personalised ads and ad SDK profiling on child accounts.

    DPDP Act s. 9(3)

  6. Deploy safety classifiers

    Filter harmful inputs and outputs, restricted to keeping detrimental content away from the child.

    Fourth Schedule Part B item 5

  7. Document a detrimental-effect safety case

    Record risks, controls and escalation for child well-being.

    DPDP Act s. 9(2)

  8. Store consent evidence, not ID documents

    Keep verification method, timestamp and notice version without retaining copies of identity documents.

    Rule 10; s. 6(1)

Key dates

  1. DPDP Rules notified

  2. Rules 10 and 12 and section 9 duties apply

Answers

Frequently asked questions

Can an AI chatbot remember a child's past conversations under DPDP?

Long-term memory that builds a profile of a child's behaviour risks breaching section 9(3), which bans tracking and behavioural monitoring of children. Session-only memory is far easier to defend. If persistent memory is essential, restrict it to a clear purpose, make it visible to parents, and take advice on whether any Fourth Schedule condition applies.

Do teenagers count as children under India's DPDP Act?

Yes. The DPDP Act defines a child as an individual who has not completed eighteen years. A 16 or 17-year-old using an AI app is a child, so verifiable parental consent and the section 9 restrictions on tracking, monitoring and targeted ads apply from 13 May 2027.

How do I verify a parent under Rule 10?

Check that the person identifying as the parent is an identifiable adult, using reliable identity and age details you already hold, details she voluntarily provides, or a virtual token issued by an authorised entity, including through a Digital Locker service provider. Record the method and outcome rather than storing identity documents.

Are AI safety filters allowed for children's accounts?

Yes. Fourth Schedule Part B item 5 disapplies sections 9(1) and 9(3) for processing to ensure information, services or advertisements likely to harm a child are not accessible to her, restricted to the extent necessary. Use that processing only for filtering, not for broader profiling.

What is the penalty for violating children's data rules under DPDP?

Non-fulfilment of the additional obligations relating to children under section 9 carries a penalty of up to ₹200 crore per instance. The Board decides the amount after inquiry, considering the nature, gravity and duration of the breach, type of data, repetition and mitigation.

Can an AI tutor adapt to a child's learning level?

Within-session adaptation to the current question is low-risk. Cross-session learning profiles are behavioural monitoring. Educational institutions are exempt from section 9(3) for tracking and monitoring for educational activities under Fourth Schedule Part A item 3; a consumer app outside an institution generally is not. Section 9(2) still applies.

Legal references and sources

  • Digital Personal Data Protection Act, 2023 — Section 9(1)–(5) (official text)
  • Digital Personal Data Protection Rules, 2025 — Rule 10 (official text)
  • Digital Personal Data Protection Rules, 2025 — Rule 12 and Fourth Schedule (official text)
  • DigiLocker — Digital Locker service provider (Rule 10(2)(b)) (official text)

This guide explains the law for engineering and business teams. It is not legal advice; confirm how it applies to your organisation with qualified counsel, and check the Gazette for later notifications or amendments.

About the author

Suhas Bhairav

Suhas Bhairav

Systems architect and applied AI practitioner who designs production AI systems for Indian and global companies, with a focus on privacy-by-design under the DPDP Act and the EU AI Act.