India's AI governance approach favours institutions and guidelines over legislation. Critically evaluate its adequacy in the age of agentic and open-source AI.
New · District CollectorCan you hold the district for 72 hours?Floods, a heatwave, an outbreak, a grievance queue. Take the decisions, see the consequences.Play →MeitY's India AI Governance Guidelines, issued under the IndiaAI Mission, set up a whole-of-government model of three bodies instead of a standalone AI Act: an AI Governance Group (AIGG), a Technology & Policy Expert Committee (TPEC) and an AI Safety Institute (AISI) [1]. The model can adapt quickly, but agentic and open-weight AI show where it cannot enforce.
Strengths of the institution-led model
- Agility: guidelines can change as fast as models do. A statute could be out of date by the time it is notified.
- Innovation-friendly: India's young AI ecosystem, still building compute under the IndiaAI Mission, avoids heavy compliance costs [1].
- Technical home: the AISI gives India a body for testing, standards and safety research [1].
- Existing laws fill gaps: the IT Act 2000 and the DPDP Act 2023 already cover cyber offences and data misuse.
- Diplomatic leverage: the soft-law approach let India win broad agreement for the India AI Impact Summit Declaration (2026) [2].
Gaps exposed by agentic and open-source AI
- No binding force: the New Delhi Frontier AI Impact Commitments were signed by 13 model providers, not by governments, and carry no legal penalty [3].
- Liability vacuum: autonomous agents carry out multi-step actions, so it is unclear whether the developer, the deployer or the user is accountable. The IT Act's intermediary framework was not designed for this.
- Access problem: without a statute, the AISI cannot make foreign labs share closed frontier models for testing before release.
- Open-weight dilemma: once model weights are public, anyone can modify them. Testing can warn about a risk but cannot recall a model.
- Fragile institutions: without statutory backing, the AIGG and TPEC depend on executive will.
Verdict and way forward The model works as a foundation but not as a finished framework.
- Give the AISI targeted statutory backing: mandatory pre-deployment testing and incident reporting for high-risk and agentic systems. This would be risk-based like the EU AI Act, but narrower.
- Set liability rules for agentic AI through amendments to the IT Act.
- Seek multilateral transparency norms at the UN Global Dialogue on AI Governance, created by consensus under A/RES/79/325 [4]. Use the 40-expert Scientific Panel's evidence for risks that cross borders [5].
India's institution-led approach rightly protects flexibility for a young ecosystem, but it lacks the binding tools that autonomous and freely spread AI require. A "guidelines-plus" model would keep innovation and accountability together and advance SDG 9 and SDG 16. Under it, soft law covers most AI, hard law covers high-risk agentic uses, and global norms cover open-weight releases.
Sources
- 1India AI Governance Guidelines: Enabling Safe and Trusted AI Innovation, MeitY/PIB (2026): AIGG–TPEC–AISI model, AISI's mandate, IndiaAI Mission framing
- 2MEA, India AI Impact Summit Declaration (Bilateral/Multilateral Documents): the Summit Declaration
- 3PIB, MeitY release on the India AI Impact Summit 2026 outcomes: Frontier AI Impact Commitments signed by 13 model providers
- 4UN General Assembly Resolution A/RES/79/325 (2025): Global Dialogue on AI Governance and Scientific Panel, adopted by consensus
- 5UN Independent International Scientific Panel on AI: FAQ: 40 independent experts on the Panel