Data Privacy1 min read
Listen to Article0:00 / 1:00

MeitY Proposes Mandatory AI Model Audits Under DPDPA Framework

Draft guidelines require large AI model developers to conduct bias audits, transparency reports, and data lineage documentation before deployment in the Indian market.

AI
AI Policy Correspondent·
#AI Regulation#DPDPA#Machine Learning
MeitY Proposes Mandatory AI Model Audits Under DPDPA Framework

The Ministry of Electronics and Information Technology has released draft guidelines proposing mandatory audits for AI models that process personal data of Indian citizens. The guidelines, issued under Section 36 of the DPDPA, would apply to AI models with more than 1 million users or those deployed in "high-risk" sectors including healthcare, finance, and criminal justice.

Audit requirements

Developers must conduct and publish bias audits examining the model's performance across demographic categories defined by the Census of India. Transparency reports must include training data composition, model architecture summaries, and known limitations.

Data lineage obligations

A key provision requires AI developers to maintain and disclose "data lineage documentation" — a traceable record of all personal data used in training, including its source, consent basis, and processing history.

Industry response

AI startups have raised concerns about compliance costs, estimating that a full audit cycle could cost between ₹50 lakh to ₹2 crore per model. Established tech companies have been more measured, noting that similar requirements already exist under the EU AI Act.

Timeline

The public consultation period closes on June 30, 2026, with final guidelines expected by Q3 2026.