How AI training models could compromise privacy
Source: The Hindu, 3 November 2026
Summary
The editorial argues that India's legal instruments do not adequately protect the constitutional right to privacy in the age of AI. The DPDP Act, 2023 and the DPDP Rules, 2025 say nothing on AI training. They also exclude publicly available and inferred data, and they have no sensitive-data category. The author asks for targeted amendments, Rules and judicial clarification rather than blanket AI regulation or a ban on using personal data for AI.
Key facts for Prelims
- Right to privacy: upheld as a constitutional right by the Supreme Court in 2017, under Article 21.
- DPDP Act, 2023 is the general-purpose data law, and the DPDP Rules were notified in November 2025.
- The Information Technology Act, 2000 and its 2011 Rules were the main instrument before DPDP. They had a "sensitive personal data" category, and the DPDP Act does not.
- DPDP terms: Data Principal, Data Fiduciary, Data Processor, Consent Manager, Significant Data Fiduciary.
- The Data Protection Board of India enforces the Act, and TDSAT is the appellate body.
- DPDP right to erasure ("right to be forgotten") applies only to data given with consent. The Act exempts processing for research, archival or statistical purposes, and the conditions for that exemption are yet to be prescribed by the Union government.
- Meta offers EU users an opt-out from AI training. Indian users have no equivalent right.
Data and details
- DPDP Act passed in 2023, Rules notified in November 2025. Some provisions on consent and safety become effective only in 2027, and the substantive provisions take full effect in May 2027.
- IT Act regime: criminal penalty for unauthorised disclosure capped at three years' imprisonment or a ₹5 lakh fine. It applied only to listed sensitive data (financial, medical or health details, sexual orientation, biometric information, etc.).
- India is the largest market in the world for Facebook and Instagram. Meta, Amazon and Google use user data to train AI models.
- Under the IT Act regime, a mere assertion and claim was sufficient to record "adequacy" for export of sensitive personal data. The transfer also had to be necessary for a lawful contract or consented to.
- DPDP puts passwords and mental and physical health conditions in the same bucket as name, marital status, city and shopping interests.
- Notably absent from the Act:
- Any engagement with AI models or training data.
- A right to object to AI training on public data, even after May 2027.
- Clarity on whether erasure extends to inferred data.
- Clarity on whether AI training counts as "research" for the exemption.
- The Act does not apply to publicly available data, per the author. "Inferred" data, such as emotions, mobility and behavioural patterns, is also outside the Act's protection.
- Studies cited: querying an AI model with "anonymised" data can extract memorised personal details and be used to build individual profiles.
- Author's reform list:
- Apply DPDP to AI-training data.
- Give a right to object.
- Protect scraped public data and inferred data.
- Make companies responsible for informing users of every entity holding their data.
- Clarify the research exemption.
- Create a sensitive-data category.
- Author's caution: do not ban personal data in AI training, since that could worsen inclusion, efficiency, discrimination and bias outcomes for Indians.
Static–dynamic linkage
Privacy was recognised as a fundamental right by the Supreme Court in 2017 (the Puttaswamy judgment). It flows from Article 21 (life and personal liberty) and is protected as an intrinsic part of dignity and liberty. A fundamental right needs an enabling statute that defines who may use personal data, for what purpose, and with what remedy. The news is an instance of the gap between a constitutional guarantee and its statutory implementation. The gap is widest where new technology (generative AI trained on scraped data) outpaces the law. It also shows the policy trade-off between innovation and rights.
Syllabus mapping: GS2 – Polity: Fundamental Rights (Article 21, right to privacy) and statutory regulatory bodies | GS3 – S&T: IT & AI, emerging-technology policy | GS2 – Governance: e-governance and data protection
Background primer
Digital Personal Data Protection (DPDP) Act, 2023
- What it is:
- India's general-purpose law on processing digital personal data. The nodal ministry is the Ministry of Electronics and Information Technology.
- Founded:
- Parliament enacted it on August 11, 2023, and it needed operational rules to become functional. The 2017 privacy judgment is the constitutional backdrop.
- Mandate:
- The framework rests on seven principles: consent and transparency, purpose limitation, data minimisation, accuracy, storage limitation, security safeguards, and accountability. Individuals may access, correct, update or erase their data and nominate someone to exercise these rights. Fiduciaries must respond to such requests within 90 days.
- Why it matters:
- balance individual privacy with the need to process data for lawful purposes, innovation and economic growth.
Information Technology Act, 2000 and 2011 Rules (the pre-DPDP regime)
- What it is:
- the earlier main instrument for personal-data management. It regulated "sensitive personal data" (financial, medical/health, sexual orientation, biometric and similar categories listed in the Rules), as described in the article.
Additional context
- The DPDP Rules, 2025 provide for phased implementation over an eighteen-month transition period.
- Phases: Phase 1 sets up the Data Protection Board, composed of a Chairperson and four Members chosen by a Search-cum-Selection Committee. Phase 2, within one year, covers registration and functioning of Consent Managers. Phase 3, within eighteen months, covers Data Fiduciaries' compliance obligations including data principal rights, security safeguards and breach notification.
- Consent Managers must be India-based companies.
- Penalties under the Act go up to ₹250 crore, the maximum, for failure of reasonable security safeguards.
- The framework also updates Section 8(1)(j) of the RTI Act to reconcile privacy with transparency.
Mains dimensions
- Regressive classification: the IT Act separated sensitive from other personal data. DPDP's single bucket gives passwords and health data the same protection as shopping interests. The author argues this weakens protection for categories that were better protected earlier.
- AI-specific harms:
- Models can regurgitate memorised training data.
- "Anonymised" datasets can be reverse-queried to re-identify people.
- Profiling can happen without consent, which erodes the Article 21 guarantee.
- Timing gap: consent obligations apply only from 2027, so companies need not fully comply now. Even afterwards there is no right to object to public-data scraping. Meta's EU opt-out shows the contrast.
- Exemption loophole: the right to erasure is excluded for research, archival and statistical processing. The conditions are unprescribed, and AI training is often framed as "research", so it could escape the Act entirely.
- Innovation vs rights: blanket regulation risks slowing AI and growth. The author prefers "low-hanging" fixes through amendment, Rules and judicial rulings. Banning data use or export could worsen bias and exclusion for Indian populations.
- Strategic angle: India is positioned as a global AI "use case hub", and the more privacy gaps are delayed, the weaker the constitutional right becomes.
- Way forward: apply DPDP to AI training, add an opt-out, protect scraped and inferred data, restore a sensitive-data tier, shift the burden of erasure to companies, and clarify the research exemption.
Possible Mains question
**GS2/GS3, 15 marks:** "The right to privacy is a fundamental right, yet India's data protection framework has gaps in the age of Artificial Intelligence. Examine the adequacy of the Digital Personal Data Protection Act, 2023 in regulating the use of personal data for AI training and suggest measures."
Related topics
- Puttaswamy judgment (2017) and the tests of legality, necessity and proportionality
- Article 21 and the evolving right to dignity
- RTI Act, 2005 and Section 8(1)(j) vs privacy
- Data localisation and cross-border data flows
- AI governance in India (IndiaAI Mission, global approaches such as the EU AI Act)
- Cyber security and the IT Act, 2000
Topics: GS2Polity
All current affairs of 3 November 2026 → · UPSC Current Affairs