Bronnen / Blog Articles / For organisations

India's DPDP Rules and AI Training Data: What Enterprise Teams Should Prepare Before May 2027

India notified the Digital Personal Data Protection (DPDP) Rules in November 2025, and the framework is now rolling out in phases. Legal commentary puts the core obligations for data fiduciaries, the organisations that decide why and how personal data is processed, at roughly 18 months after notification, which lands in May 2027.

For teams that train or evaluate AI models on data from Indian residents, that leaves a short runway. This guide explains what the rules mean for training data, where the open questions are, and what to ask of any vendor that collects data for you. It is an operational overview, not legal advice, so confirm dates and obligations with counsel.

Where the DPDP rollout stands in November 2026

The rollout is staggered. Provisions that set up the Data Protection Board took effect on notification. A second group, covering consent managers (registered intermediaries that let individuals manage their consents), was scheduled for around 12 months after notification, although published summaries differ on the exact scope and date. The largest group, covering notice, consent, security safeguards, breach reporting and data retention, applies at around 18 months.

Sources also differ by a day or two on the exact dates, so use the official Gazette notification as your reference. The practical point is the same either way: the period for building compliant consent and record-keeping processes is already more than half gone.

Why AI training data is the pressure point

The DPDP Act does not create a separate regime for AI. It applies general data protection duties to anyone processing digital personal data, and Indian law firms note that AI developers will usually be treated as data fiduciaries because they set the purpose and means of processing.

Two features matter for training data. First, commentators point out that the Act has no broad legitimate-interest basis and no general research exemption for commercial model training, which is different from the GDPR in Europe. Consent, tied to a specific stated purpose, carries most of the weight. Second, the rules require clear notices and the ability to withdraw consent, which is hard to honour if you cannot trace which records came from which person.

  1. Voice recordings, face images and video of identifiable people are personal data, even when collected only to train a model.

  2. Text written by contributors can contain personal data about them or about third parties.

  3. Data scraped or bought without a clear consent trail is the hardest category to defend.

  4. Data about children carries stricter conditions, including verifiable parental consent.

What to document for every dataset

Most of the work is record-keeping. If a regulator, an auditor or a customer asks how a dataset was built, you should be able to answer from documents instead of from memory.

  1. The notice each contributor saw, in the language they saw it in, with a version number and date.

  2. The specific purpose consented to, for example speech recognition training, and whether it covers reuse for other models.

  3. A consent record linked to each recording or file, so a withdrawal can be actioned.

  4. Where the data is stored and which countries it is transferred to.

  5. Retention periods and proof of deletion when they expire.

  6. The security controls applied by you and by each vendor in the chain.

Latin America is moving in the same direction

India is not the only large market tightening the rules. Brazil's AI bill, PL 2338/2023, passed the Senate in December 2024 and remained with the Chamber of Deputies in early November 2026, with no vote scheduled according to recent reporting. Coverage of the current text says it would require AI developers to negotiate remuneration with rights holders whose works are used for training. The text can still change, and Brazil's existing data protection law, the LGPD, already applies to personal data in training sets.

The common thread across India, Brazil and the EU is provenance. Regulators want to know where training data came from and on what terms. A team that builds that record once can reuse it in each market.

Questions to ask your data vendor now

If a vendor collects or labels data for you, your exposure includes their process. Ask these questions before the next statement of work, not after a complaint.

  1. Can you show the consent notice and consent record for a sample of files in our dataset?

  2. How do you verify contributor identity and age?

  3. How would you find and remove one contributor's data if they withdraw consent?

  4. Where are contributors located, and where is the data processed?

  5. What audit trail exists for quality review and for changes to annotation guidelines?

Corpshore AI, the company behind the Jwuma contributor platform, runs consent capture, identity verification and quality-review audit trails as part of its managed data services, so clients receive that documentation with the data. Jwuma contributors work in 35+ languages across 80+ countries, including Hindi, Bengali, Tamil, Spanish and Portuguese.

Veelgestelde vragen

Does India's DPDP Act apply to AI training data?

Yes, when the training data includes digital personal data about people in India. The Act has no separate AI chapter, so general duties such as notice, consent, security and deletion apply to training datasets.

When do the DPDP Rules fully take effect?

The rules were notified in November 2025 with a phased rollout. Legal summaries place the core data fiduciary obligations about 18 months later, in May 2027. Confirm exact dates against the official Gazette notification.

Can I train an AI model on Indian personal data without consent?

Commentators note the DPDP Act has no broad legitimate-interest or general research basis for commercial AI training, so consent for a stated purpose is the main route. Take legal advice for your specific case.

What is a data fiduciary under the DPDP Act?

A data fiduciary is the organisation that decides the purpose and means of processing personal data. AI developers that choose what data to collect and why will usually fall into this role.

Has Brazil passed its AI law?

Not yet. Brazil's AI bill, PL 2338/2023, passed the Senate in December 2024 and was still before the Chamber of Deputies in early November 2026, with no vote scheduled.

Talk to Corpshore AI about consent-documented training data →

Published 2026-11-19 by Jwuma, operated by Corpshore AI. This piece is written for organisations sourcing AI data work. Visit client.corpshore.ai om een programma te bespreken.