New UPSC Foundation, Optional and TGPSC/APPSC batches are open — book your counselling slot.Explore coursesGroup 2 crash course ₹9,999Today's Current AffairsToday's Daily QuizPlaces in NewsCall 98804 87071

Economic Survey 2026 · Chapter 14 · Topic 12 of 15

Regulating AI Firms

Core framework principles

Governance

Quick answer

Economic Survey 2025-26 proposes transparency-centred AI regulation under the DPDP Act, 2023 framework, using public access incentives rather than model-location controls.

Regulating AI firms means making developers and deployers answerable for data use, model limitations and harms throughout a system’s life cycle. The Survey favours transparent conduct, proportionate duties and access incentives over attempts to control the physical location of model parameters.

Source: paragraphs 14.76–14.80 and Box XIV.6; these are recommendations, not a newly enacted AI licensing code.

Concept diagram

Regulate conduct, preserve useful innovation

A regulator’s checklist connects proportionate obligations to oversight and enforceable responsibility.

  • Obligations should reflect context and risks rather than treating all uses alike.
  • Clear accountability and credible enforcement are as important as written rules.

Basics: what must be governed?

Model parameters

  • Weights are learned numerical parameters, not the same thing as a training dataset.
  • Distributed training and modular deployment weaken location-based controls.

Impact assessment

  • An impact assessment examines likely consequences in a deployment context.
  • The Survey calls for publication of assessments for high-risk deployments.

Post-deployment duty

  • Monitoring checks real-world performance after release.
  • Harm reporting creates a feedback channel when laboratory tests miss failures.

The proposed compliance chain

StageFirm obligationPublic purpose
Training inputsMaintain clear dataset-provenance recordsMake origin and downstream uses traceable
Model descriptionPublish standardised documentation on inputs and limitationsReduce information asymmetry
High-risk deploymentUndertake and publish impact assessmentsExpose consequential risks to public review
After deploymentMonitor systems and report harmsEnable correction as conditions change
Model-weight exportsRestrict only narrowly defined high-risk categories with demonstrable harmsPreserve ordinary openness while retaining targeted intervention

Box XIV.6 in context

  • The DPDP Act, 2023 is the statutory foundation discussed for evolving data governance; the proposed AI duties must not be presented as all already enacted.
  • Over 100 crore broadband-connected people form the market context in paragraph 14.61, not a count of regulated AI firms.
  • Certified domestic compute or data environments are voluntary; reduced audit burdens and faster clearances make participation attractive.

Access is the State’s strongest practical lever

  1. Set accountability conditions

    Specify documentation, oversight and proportionate value-retention obligations.

  2. Attach public benefits

    Link eligibility for government datasets, national missions, sandboxes and procurement to compliance.

  3. Reward trusted environments

    Offer regulatory facilitation and preferential national-programme access without making domestic participation compulsory.

Incentives versus indiscriminate mandates

Positive incentives

  • Firms choose certified domestic environments.
  • Public opportunities reward accountable behaviour.

Blanket physical controls

  • A universal weights-export ban would exceed the proposed approach.
  • Controlling model location is impractical across distributed pipelines.
VS

Narrow restrictions remain possible where demonstrable high-risk harms justify them.

Exam traps: distinguish the claim from the inference

Statements examiners like to twist — learn the correct version.

  • Model weights and personal training records are interchangeable.

    Weights are learned parameters; provenance records concern the data used in development.

  • Voluntary certification removes all accountability.

    Participation can be voluntary while public-benefit eligibility remains conditional on compliance.

  • The Survey recommends banning every model-weight export.

    It reserves restrictions for narrowly defined high-risk categories with demonstrable harms.

Prelims focus

  • Identify the four accountability tools: provenance, documentation, impact assessments and monitoring.
  • Public procurement and dataset access are incentive levers, not synonyms for blanket licensing.

Mains angle

  • Assess how access-based incentives can regulate powerful AI firms without unnecessarily expanding statutory controls.
  • Discuss transparency as a response to information asymmetry between AI developers, users and regulators.

Flip & recall

Answer in your head first, then tap the card to check.

Practice MCQs

Original Prelims-style questions with explanations.

0 / 3 correct · 3 left

  1. Q1.Which is the Survey’s preferred general approach?

  2. Q2.What is the proposed status of certified domestic compute participation?

  3. Q3.Which claim is consistent with the model-weight export recommendation?

Free UPSC daily current affairs quiz — 10 questions, new every day at 8 am IST.

Call nowWhatsApp