DPDP Rules 2025Updated 22 November 20269 min read

Algorithmic Due Diligence for Significant Data Fiduciaries

Rule 13(3) requires Significant Data Fiduciaries to verify that technical measures, including algorithmic software, are not likely to pose a risk to Data Principals' rights. Here is what that means in practice.

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

Provision
Rule 13(3), DPDP Rules, 2025
Who
Significant Data Fiduciaries (once notified)
Scope
Technical measures including algorithmic software handling personal data
Standard
Not likely to pose a risk to the rights of Data Principals
In force from
13 May 2027
Maximum penalty
Up to ₹150 crore per instance (SDF obligations)

Quick answer

Rule 13(3) of the DPDP Rules, 2025 requires a Significant Data Fiduciary to observe due diligence to verify that technical measures, including algorithmic software, adopted for hosting, display, uploading, modification, publishing, transmission, storage, updating or sharing of personal data are not likely to pose a risk to the rights of Data Principals. It applies from 13 May 2027, once an entity is notified as an SDF.

  • Rule 13(3) is India's first explicit algorithmic due diligence duty for personal data, and it binds only Significant Data Fiduciaries.
  • It covers technical measures including algorithmic software used across nine listed processing activities, from hosting to sharing.
  • The test is whether the software is likely to pose a risk to Data Principals' rights, not whether it is perfect.
  • The Rules prescribe no method, so documented testing, monitoring and change control are the practical evidence.
  • The duty applies from 13 May 2027 and only after SDF notification.

Who this is for: Social media and content platforms, ML and data science leads, Data Protection Officers of likely SDFs, Trust and safety teams, AI product managers

What does Rule 13(3) actually require?

Rule 13(3) reads: a Significant Data Fiduciary "shall observe due diligence to verify that technical measures including algorithmic software adopted by it for hosting, display, uploading, modification, publishing, transmission, storage, updating or sharing of personal data processed by it are not likely to pose a risk to the rights of Data Principals."

Broken down, the rule has four elements:

  1. Actor — only a Significant Data Fiduciary (SDF). No SDF has been notified yet.
  2. Object — "technical measures including algorithmic software". Algorithms are named, but the rule also covers non-algorithmic technical measures such as storage systems and sharing pipelines.
  3. Activities — nine verbs: hosting, display, uploading, modification, publishing, transmission, storage, updating and sharing of personal data. Recommendation feeds, ranking, search, moderation, upload filters and data-sharing APIs all fall inside.
  4. Standard — the SDF must verify that these measures are "not likely to pose a risk to the rights of Data Principals". It is a likelihood test, not a guarantee.

The "rights of Data Principals" are the rights under the Act: access (s.11), correction and erasure (s.12), grievance redressal (s.13) and nomination (s.14), alongside the protections built into consent, notice and children's provisions.

The wording echoes the vocabulary of the IT Rules, which also talk about hosting, display, uploading, modification, publishing, transmission, storage, updating and sharing. That makes large intermediaries the most obvious audience, but any SDF running automated systems over personal data is in scope. Rule 13 comes into force on 13 May 2027.

Which systems fall within algorithmic due diligence?

Because Rule 13(3) is keyed to processing activities rather than technology labels, the simplest way to scope it is to map each activity to the systems that perform it.

Activity in Rule 13(3)Typical systems
Hosting, storageDatabases, object stores, data lakes, vector stores
Display, publishingFeeds, ranking and recommendation models, search
Uploading, modificationUpload pipelines, image/video processing, content filters
UpdatingProfile enrichment, entity resolution, automated data correction
Transmission, sharingData-sharing APIs, ad-tech integrations, partner exports

Some categories deserve special attention:

  • Profiling and inference. Models that derive new attributes about people from behaviour can affect accuracy obligations under section 8(3) where data is used for decisions.
  • Generative AI. LLM-based features that summarise, generate or retrieve personal data, including retrieval-augmented generation over user records, involve storage, modification and display.
  • Children. Any model that could deliver tracking, behavioural monitoring or targeted advertising to children risks breaching section 9(3) unless a Fourth Schedule exemption applies.
  • Erasure paths. Systems that resurrect deleted data from caches, embeddings or feature stores undermine the erasure right in section 12.

Third-party models and processors are still in scope. Section 8(1) keeps the fiduciary responsible for processing done on its behalf, so a vendor's recommendation API or hosted model needs the same scrutiny as in-house code.

How to perform algorithmic due diligence in practice

The Rules do not prescribe a methodology, so an SDF has to design one that it can defend before the Data Protection Board. A workable programme has five layers.

1. Inventory

Maintain a register of every model and automated decision system touching personal data: owner, purpose, inputs, outputs, training data sources, vendor, and the Rule 13(3) activities it performs.

2. Pre-deployment assessment

Before launch or material change, assess risks to Data Principals' rights:

  • Does the system use data beyond the purpose in the Rule 3 notice?
  • Can its outputs be explained well enough to answer an access request under section 11?
  • Can a correction or erasure under section 12 propagate into features, embeddings and training sets?
  • Could it target or profile children?

3. Testing

Run documented tests for accuracy, leakage of personal data in outputs, re-identification, and unfair outcomes across user groups. Keep test datasets and results versioned.

4. Monitoring

Log model versions, inputs and decisions with retention aligned to Rule 6(1)(e) and Rule 8(3), so incidents can be investigated. Monitor drift and complaint patterns from the grievance system.

5. Change control and sign-off

Tie model releases to an approval step that records the due diligence outcome. Feed results into the annual DPIA so significant observations reach the Board.

What evidence will satisfy the Board?

"Due diligence to verify" is a process obligation. If the Board inquires, the SDF will need to show that it actually verified, not merely that it intended to. Useful evidence includes:

  • Model cards or system cards describing purpose, data, limitations and known risks.
  • Assessment records for each launch and material change, with named reviewers and dates.
  • Test reports showing what was tested, against which thresholds, and what happened when a threshold was missed.
  • Production logs of model versions and decisions, retained for at least one year under Rule 8(3).
  • Vendor due diligence for third-party models and APIs, including contract clauses on security safeguards under Rule 6(1)(f).
  • Complaint linkage showing that grievances about automated outcomes are triaged and fed back into model review.
  • Remediation tickets that close the loop on identified risks.

The audit required by Rule 13(1) is the natural place for an independent check of this evidence. An auditor sampling five model releases and finding signed assessments, tests and monitoring for each is a far stronger position than a policy document alone.

Keep the documentation proportionate. A ranking model serving crores of users needs deeper analysis than an internal deduplication script, and the register should record why a lighter review was chosen.

How does Rule 13(3) interact with other AI rules?

Rule 13(3) sits alongside other Indian and international expectations on algorithmic systems.

  • MeitY India AI Governance Guidelines (November 2025) are principle-based and voluntary. Their sutras on accountability, fairness and being understandable by design align well with Rule 13(3) evidence.
  • IT Rules already impose due diligence on intermediaries under Rule 3(1)(b), with safe harbour under section 79 of the IT Act at stake. Large intermediaries likely to be SDFs should run one combined assessment rather than two.
  • Sector regulators. RBI's FREE-AI committee report recommends AI inventories, risk classification and model validation for regulated entities, though these bind only once RBI issues directions. SEBI regulation 16C makes regulated persons solely responsible for outputs of AI tools they use.
  • EU AI Act. Indian companies serving the EU may already maintain risk management and documentation for high-risk or GPAI systems. That material can be reused, but Rule 13(3) is framed around rights of Data Principals under DPDP, not EU risk tiers.

The common thread is a reliable model inventory and change log. Build that once and map it to each framework.

For the remaining SDF duties, see Significant Data Fiduciary obligations.

Common mistakes and what to do before May 2027

Mistakes to avoid

  • Reading "algorithmic" too narrowly. Rule 13(3) says "technical measures including algorithmic software", so storage and sharing systems count too.
  • Ignoring vendor models. Outsourcing a model does not outsource accountability under section 8(1).
  • One-time reviews. Verification at launch is not enough if the model is retrained weekly. Tie due diligence to change control.
  • No link to rights handling. If a correction or erasure request cannot reach derived data and embeddings, the system is likely to pose a risk to Data Principals' rights.
  • Undocumented trade-offs. If a known risk is accepted, record why and who decided.

Before 13 May 2027

  1. Decide whether you are a plausible SDF candidate using the section 10(1) factors.
  2. Build the model and automated-system inventory.
  3. Write an assessment template keyed to the nine Rule 13(3) activities.
  4. Pilot it on your two or three highest-impact systems.
  5. Wire model version logging into your one-year retention store.
  6. Agree with your auditor how Rule 13(3) evidence will be sampled.

Organisations that are not SDF candidates can adopt a slimmer version: it costs little and strengthens your Rule 6 security posture.

Further reading on suhasbhairav.com

Before 13 May 2027

Compliance checklist

  1. Build an algorithmic systems register

    List every model and automated system touching personal data with owner, purpose, data sources and Rule 13(3) activities.

    Rule 13(3)

  2. Define a pre-deployment assessment

    Require a documented rights-risk review before launch and before every material change.

    Rule 13(3)

  3. Test for leakage and accuracy

    Test outputs for personal data leakage, re-identification and inaccurate inferences, and keep results versioned.

    Section 8(3); Rule 13(3)

  4. Propagate rights requests

    Ensure correction and erasure flow into feature stores, embeddings and caches.

    Section 12; Rule 14

  5. Log model versions and decisions

    Retain logs for at least one year so incidents and complaints can be investigated.

    Rule 6(1)(e); Rule 8(3)

  6. Review vendor models

    Apply the same due diligence to third-party models and APIs and add safeguard clauses to contracts.

    Section 8(1); Rule 6(1)(f)

  7. Feed results into the DPIA and audit

    Include algorithmic findings in the annual DPIA and make the evidence available to the auditor.

    Rule 13(1)–(2)

Key dates

  1. DPDP Rules, 2025 published

  2. Rule 13(3) algorithmic due diligence comes into force for notified SDFs

Answers

Frequently asked questions

What is algorithmic due diligence under the DPDP Rules?

It is the duty in Rule 13(3) for a Significant Data Fiduciary to verify that technical measures, including algorithmic software, used for hosting, display, uploading, modification, publishing, transmission, storage, updating or sharing of personal data are not likely to pose a risk to the rights of Data Principals. It applies from 13 May 2027.

Does Rule 13(3) apply to every company using AI?

No. Rule 13(3) binds only Significant Data Fiduciaries notified by the Central Government under section 10 of the DPDP Act, and none has been notified yet. Other companies using AI still have general duties such as reasonable security safeguards and accuracy where data is used for decisions.

Does the DPDP Act ban automated decision-making?

No. The DPDP Act has no standalone right to object to automated decisions. Instead, Rule 13(3) requires Significant Data Fiduciaries to verify that algorithmic software is not likely to harm Data Principals' rights, and section 8(3) requires completeness, accuracy and consistency where personal data is used to make decisions.

Are third-party AI models covered by Rule 13(3)?

Yes, in practice. Rule 13(3) refers to technical measures adopted by the Significant Data Fiduciary, and section 8(1) keeps the fiduciary responsible for processing carried out by its processors. A vendor model or hosted API used on personal data should go through the same assessment as an in-house model.

Is there a prescribed method for algorithmic due diligence?

No. The Rules set the standard, not the method. A defensible approach combines a system inventory, pre-deployment risk assessment, documented testing, production monitoring with one-year logs, and change control, with results summarised in the annual DPIA and checked during the audit.

What is the penalty for failing algorithmic due diligence?

Rule 13(3) is an additional obligation of a Significant Data Fiduciary, and breach of SDF obligations can attract a penalty of up to ₹150 crore under the Schedule to the DPDP Act. The Board fixes the amount after inquiry, considering nature, gravity, duration and mitigation.

Legal references and sources

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.