Is India Mandating Local Storage of AI Models Under the DPDP Act?

Key takeaways

• The DPDP Act, 2023 has no clause that mandates local storage of AI models. Its cross-border rule works by notifying countries to which transfers may be restricted.
• A localisation requirement for AI models was reported as a proposal, not a settled law. Verify the current position with MeitY and the notified rules.
• If you build or use AI with Indian personal data, the practical duties are consent, purpose limits, security safeguards, breach reporting and respecting data principal rights.
• Sector rules can be stricter than the DPDP Act, for example RBI rules on payment data storage.
• This is general information, not legal advice.

Does the DPDP Act require AI models to be stored in India?

No, not in the text of the Act. The Digital Personal Data Protection Act, 2023 regulates how organisations (data fiduciaries) process the digital personal data of individuals (data principals). It does not contain a clause that requires AI models to be hosted in India. A news report in 2025, which this article was originally based on, said the government was considering a policy to require local storage of AI models and that the rules under the Act were expected within weeks. A consideration is not a mandate. Always check the Act and rules on the MeitY website.

What does the Act say about cross-border data?

Section 16 of the Act takes a "negative list" approach: personal data may be transferred outside India except to countries or territories the Central Government notifies as restricted. Other Indian laws that impose stricter requirements continue to apply, so if a sector regulator demands local storage, that still holds. A well-known example is the RBI's 2018 directive on storage of payment system data in India. Check your sector rules.

What is the difference between data localisation and model localisation?

Why would a government want AI models stored locally?

The arguments usually given are control and enforcement (easier access for regulators), protection from foreign legal demands, reduced risk that prompts and training data leave the country, and building domestic capacity. The counter-arguments are cost and capacity, slower access to the latest models, and the fact that location does not by itself guarantee security. Both views exist, and the policy outcome is uncertain.

What security risks do AI systems create for personal data?

• Prompts and outputs: staff paste customer data into public AI tools.
• Training data: models can memorise and reveal personal data if it was in the training set.
• Third-party processing: data goes to a vendor's cloud, sometimes outside India.
• Access control and logging: weak controls on model endpoints and vector stores.

The OWASP project maintains guidance on large language model application risks, and CERT-In publishes advisories and directions at cert-in.org.in.

What should organisations do now?

1. Map your data. Which AI systems touch personal data, and where is it processed and stored?
2. Know your lawful basis and notices. Consent and purpose limits under the Act apply to AI uses too.
3. Review vendors. Contracts, data processing terms, location of processing, deletion and breach duties.
4. Set an AI use policy. What staff may and may not paste into AI tools.
5. Secure the stack: encryption, access control, logging and incident response, with a plan for breach reporting.
6. Track the rules. Review MeitY notifications and your sector regulator, and take legal advice before major decisions.

What penalties does the Act provide?

The Act sets monetary penalties through its Schedule, with the highest tier applying to failures to take reasonable security safeguards to prevent a personal data breach. Read the Schedule in the Act for the current amounts and conditions rather than relying on a summary.

How does this compare with GDPR?

GDPR restricts transfers through adequacy decisions and safeguards such as standard contractual clauses, while the DPDP Act uses a notified-country approach. Both require lawful processing and security. See our comparison of GDPR and CCPA for another view of privacy regimes.

Next steps

To understand the compliance side in depth, see WebAsha's CISM training or CISA training. Related reading: security versus privacy.