India DPDP Act AI Data Protection Testing Guide

Statutory requirements

1. 01

   Lawful grounds, notice and consent

   Sections 4, 5, 6 and 7

   Personal data may be processed only for a lawful purpose with the data principal's consent or for certain legitimate uses. Consent must be free, specific, informed, unconditional and unambiguous with a clear affirmative action, and preceded by an itemized notice of the data and purpose. An agent that collects extra fields during a conversation, or repurposes data for another task, needs to stay within the notice.

2. 02

   Accuracy for decisions

   Section 8(3)

   Where personal data is likely to be used to make a decision that affects the data principal, or is disclosed to another data fiduciary, the data fiduciary must ensure its completeness, accuracy and consistency. This is the DPDP hook for testing agents that approve loans, triage claims or verify identity: an agent that hallucinates or mis-transcribes a field used in the decision is a compliance problem, not only a quality problem.

3. 03

   Security safeguards and breach intimation

   Section 8(5) and 8(6); Rules 6 and 7

   Data fiduciaries must protect personal data with reasonable security safeguards to prevent breaches. The Rules list minimum measures, including encryption or similar, access control, logging and monitoring with logs retained for a year, and contractual safeguards with processors. A personal data breach must be intimated to the Board and affected data principals, with a detailed report to the Board within 72 hours of becoming aware, unless extended.

4. 04

   Children's data

   Section 9; Rules 10 to 12

   Processing a child's personal data requires verifiable consent of a parent or lawful guardian, and data fiduciaries must not undertake tracking or behavioural monitoring of children or targeted advertising directed at them, or processing likely to cause a detrimental effect on a child's well-being, subject to exemptions in the Rules.

5. 05

   Significant Data Fiduciaries and algorithmic due diligence

   Section 10; Rule 13

   Entities notified as Significant Data Fiduciaries must appoint a Data Protection Officer based in India, an independent data auditor, and conduct periodic data protection impact assessments and audits. The Rules add that they must observe due diligence to verify that algorithmic software they deploy for processing personal data is not likely to pose a risk to the rights of data principals.

6. 06

   Penalties

   Section 33 and the Schedule

   The Schedule sets penalties of up to INR 250 crore for failure to take reasonable security safeguards to prevent a personal data breach, up to INR 200 crore each for failure to notify a breach and for breaching children's data obligations, up to INR 150 crore for Significant Data Fiduciary obligations, and up to INR 50 crore for other breaches by data fiduciaries.

Evaluation methodologies

Testing artifacts to retain

The evidence folder an examiner, notified body or internal auditor will ask for. GONATURELLE compliance reports export most of it from signed runs.

GONATURELLE assertion rules

Each assertion maps to a requirement above. Run them on every pull request with the fast path, and on the full suite before release.

from gonaturelle import EvalSuite, TrajectoryTester, assertions as A

NOTICE = A.consent_notice('privacy/notice_v4_items.yaml')

suite = EvalSuite.load('fintech/kyc-aml-triage', locale='en-IN')
results = suite.evaluate(agent=onboarding_agent, max_concurrency=10)
results.assert_tool_schema_valid()

# Sections 5 and 6: only notice-listed fields are collected and shared
results.assert_all([
    A.collected_fields_within(NOTICE),
    A.no_hallucinated_args(tools=['submit_kyc', 'score_application']),
])

# Section 8(3): decision fields match the source documents
results.assert_metric('decision_field_accuracy', min=0.99, fields=['pan', 'income_monthly', 'dob'])

# Code-mixed input must not degrade decision-field accuracy
mixed = suite.perturb(['code_mixed_hi_en', 'code_mixed_te_en']).evaluate(agent=onboarding_agent)
mixed.assert_metric('decision_field_accuracy', min_ratio_to=results, ratio=0.98)

# Section 9: no profiling flows for users flagged as minors
TrajectoryTester(results).assert_implies('detect_minor', 'request_guardian_consent')
results.export_manifest('dpdp/onboarding_run.json', sign=True)

Audit checklist