AI ethics in India — a working framework for 2026 boards

AI ethics

The principles, policies and review processes that govern whether and how an organisation deploys AI — distinct from compliance (which is what law and certification require) and safety (which is operational harm prevention).

The five pillars, with a working board statement for each

1. Fairness

Working statement: "We will measure model outcomes across protected attributes including the Indian-context attributes (caste category where self-declared, religion, region, urban/rural) and remediate any disparity above threshold X before deployment, with the threshold and remediation reviewed quarterly by the AI Ethics Committee."

The threshold is yours to set. Common starting point: equal opportunity difference of less than 5 percentage points across protected groups. The point is not the number, it is having a number.

2. Accountability

Working statement: "A named executive (Chief AI Officer or Head of AI Governance) is accountable to the board for AI ethics outcomes. Each production AI system has a named system owner accountable for its ongoing ethical posture."

If you cannot fill in two names today, you do not have an AI ethics policy yet.

3. Transparency

Working statement: "Users will be told when they are interacting with an AI, what data the AI uses to make decisions about them, and how to contest those decisions. Internal stakeholders will have access to model cards and impact assessments for every production AI system."

4. Privacy

Working statement: "AI systems will process personal data only on the narrowest lawful basis under DPDP and, where applicable, GDPR. Training data provenance will be documented for every model. Deletion requests will propagate to training derivatives where technically feasible."

See our companion essay on GDPR vs DPDP for the dual-regime view.

5. Indian context

Working statement: "We commit to language inclusion (testing in at least Hindi-English code-mix), to protection of vulnerable Indian populations (rural, low-literacy, gig and informal workers), to caste/region/religion-sensitive deployment, and to alignment with the NITI Aayog #AIForAll principles."

From policy to control — the operating layer

Each principle needs to map to at least one control with a measurable output. The working pattern:

• Fairness → quarterly fairness audit, threshold-based remediation log.
• Accountability → named-owner register, board-level AI risk report quarterly.
• Transparency → disclosure UI shipped, model cards published internally.
• Privacy → DPIA register, retention schedule, deletion-propagation runbook.
• Indian context → multilingual red team, vulnerable-population impact assessment, NITI principles annual self-attestation.

The 90-day rollout that actually works

1. Week 1–2: Board ratifies the five-pillar statement.
2. Week 3–4: CAIO / Head of AI Governance appointed. System-owner register populated.
3. Week 5–8: Each pillar mapped to one or more controls. Owners assigned. Tooling chosen.
4. Week 9–12: First quarterly review held. Board sees first AI ethics report.

What an ethics framework cannot do

It cannot answer "should we build this product at all?" That is a strategy question, not an ethics question. The framework can refuse to ship a system that fails its controls; it cannot tell you whether to enter the surveillance-tech market. Keep ethics and strategy in their proper compartments.

Equally, an ethics framework cannot substitute for governance. A great policy with no governance process is decoration. A weaker policy run through a real quarterly board review will outperform it every time. The framework above is the starting kit; the governance loop is where it lives or dies. See /lab/policy for the policy-generation engine, or /consult to run the rollout with Dr. Sodhi.

Dr. Nitnem Singh Sodhi is a Lead Auditor for ISO/IEC 42001, 27001 and 27701, accredited by ANSI/ABICB since March 2025.

— Bharat NeuroTech · /dr-sodhi