LLexfolio

How to Write an AI Use Policy for Your Company in India: 9 Steps

Last updated: 12 November 2026

Short answer: No Indian law tells a company to adopt an “AI policy” by name, but a written policy is the simplest way to show that you took reasonable care under the laws that do apply: the DPDP Act, 2023 and Rules, the IT Act, the Consumer Protection Act and any rules from your sector regulator. This guide gives a practical sequence for a company with anything from ten to a few thousand staff. It assumes you want a policy that people actually follow, not a document that sits in a shared folder.

Need ready structure for a full programme? The Lexfolio Certificate Course in AI Law and Governance in India: Compliance for Businesses and Institutions is a self-paced certificate course on AI law and governance in India, written for managers, compliance teams and institutions that use AI tools. Price: Rs 599.

Step 1: Find out what is already in use

Most organisations discover that AI is already inside the business before anyone approved it. Ask each department head to list tools used for drafting, translation, meeting notes, CV sorting, design, customer replies, coding and analytics. Include features that vendors switched on inside software you already pay for, and personal accounts that staff use for work. Record, for each tool, the vendor, who uses it, what types of data go in and whether the output reaches customers.

Keep this as a living register with a named owner. It becomes the base for every later step, and it is also the first thing a regulator, auditor or client will ask to see after an incident.

Step 2: Sort tools by risk

Not every tool needs the same scrutiny. A grammar checker and an automated loan scoring model should not sit in the same bucket. The India AI Governance Guidelines (MeitY, 5 November 2025) call for responses proportionate to risk, and a simple three tier sort is enough to start.

TierTypical useMinimum controls
LowSpell check, translation of public text, internal brainstorming with no personal dataApproved tool list; no personal or confidential data
MediumDrafting customer replies, summarising meetings, marketing content, analytics on pseudonymised dataHuman review before release; vendor terms checked; logging of use
HighDecisions about people (hiring, credit, discipline, pricing), health or child data, public facing chatbots that make commitmentsWritten impact assessment, named approver, human override, testing, complaint route

Step 3: Map each use to the law that applies

Ask five questions of every medium and high tier use.

  • Personal data? If yes, the DPDP Act applies and your organisation is the Data Fiduciary. Notice, consent or another lawful ground, security safeguards and breach intimation follow. Most business duties start on 13 May 2027 under the Rules, so design for them now.
  • Children’s data? Section 9 of the DPDP Act requires verifiable parental consent and prohibits tracking, behavioural monitoring and targeted advertising directed at children, subject to the exemptions in the Rules.
  • Content for the public? The Consumer Protection Act, 2019 and the Central Consumer Protection Authority’s rules on misleading advertisements apply, and the 2026 IT Rules require labelling of permitted synthetic content on covered platforms.
  • Regulated sector? Banks, NBFCs, insurers, brokers and healthcare providers should check circulars from their own regulator. The RBI’s FREE-AI committee report of August 2025, for example, makes 26 recommendations for responsible AI in finance.
  • Impersonation risk? Section 66D of the IT Act (cheating by personation using a computer resource) carries up to three years and a fine up to Rs 1 lakh.

Step 4: Write the policy in plain language

A workable policy fits on four pages. The skeleton below is original drafting guidance, not statutory wording.

  • Purpose and scope: who it covers (employees, contractors, interns) and which tools.
  • Approved tools: a list, with a route to request a new tool. Anything not listed needs approval.
  • Data rules: what may never be entered (client secrets, unpublished financial results, personal data of customers or staff, passwords) and what may be entered after masking.
  • Human responsibility: the person who sends or relies on an output owns it. Outputs are checked for accuracy before use.
  • Transparency: when customers must be told that they are dealing with an automated system, and when content must be labelled as synthetic.
  • Prohibited uses: creating the voice or likeness of a real person without written permission, automated decisions about staff without review, uploading confidential material to unapproved tools.
  • Reporting: whom to tell about an error, leak or harmful output, with no blame for good faith reports.
  • Consequences and review date: disciplinary link and a review at least once a year.

Step 5: Check vendors before you sign

The DPDP Act makes the Data Fiduciary responsible for processing done on its behalf, regardless of what the contract says. Your contract should therefore carry the controls you cannot delegate. Put these questions to every vendor that touches personal data:

  1. Where is our data stored and who can access it?
  2. Are our inputs or outputs used to train or improve the vendor’s models, and can that be switched off?
  3. How long is data kept, and can it be deleted on request with proof?
  4. What security measures protect it, and how will we be told of a breach and how quickly?
  5. Which sub-processors are used?
  6. Who is responsible if the tool produces a harmful or infringing output, and is there an indemnity or a cap?
  7. Can we get logs showing what was processed, if a regulator or a person asks?

Step 6: Prepare a breach and incident route

Section 8(6) of the DPDP Act requires intimation of a personal data breach to the Data Protection Board and to each affected person. As summarised by SCC Online, the Rules require an initial description to the Board and a detailed report within 72 hours. Decide in advance who declares an incident, who calls the vendor, who writes to affected people and who speaks to the media. Add AI specific incidents to the same route: a chatbot giving harmful advice, a tool exposing one customer’s data to another, a deepfake of your director circulating. The Guidelines also recommend accessible grievance channels for people harmed by AI, so give customers a visible way to complain.

Step 7: Update your privacy notice and forms

Rule 3 of the DPDP Rules, effective 13 May 2027, requires notices that are clear and independent, listing the personal data collected, the purposes, and how to withdraw consent, exercise rights and complain. If an AI tool processes customer data for a purpose beyond what the notice states, update the notice before the tool goes live. For children, plan verification of a parent’s identity as an adult under Rule 10.

Step 8: Train people and test the policy

A policy that is not understood will be ignored. Run a short session for each department using that department’s own examples: HR with CV sorting, sales with email drafting, finance with invoice tools. Finish with a three question quiz and a sign-off. Then test the rules: ask a colleague to try to leak dummy data through an approved tool and see whether controls catch it.

Step 9: Review on a schedule

Put four dates in the calendar: a quarterly refresh of the tool register, a review of vendor terms at each renewal, a policy review after 13 May 2027 when the main DPDP duties begin, and a review whenever a new rule or court ruling affects you. Record each review in a short note signed by the policy owner.

Take the course to build your AI governance programme step by step. The Lexfolio Certificate Course in AI Law and Governance in India: Compliance for Businesses and Institutions is a self-paced certificate course on AI law and governance in India, written for managers, compliance teams and institutions that use AI tools. Price: Rs 599.

Frequently asked questions

Is an AI use policy mandatory in India?

No law uses that exact requirement. A written policy is a practical way to show reasonable care under the DPDP Act, the IT Act, consumer law and sector rules, and it helps with client and audit questions.

Who should own the AI policy in a mid sized company?

A senior person who can work across HR, IT, legal and operations, often the head of compliance or the chief technology officer, with sponsorship from the management. One named owner matters more than the title.

Can staff use free public chatbots for work?

Only if your policy allows it and only without personal data, client secrets or unpublished material. Many organisations permit public tools for low risk tasks and require approved enterprise accounts for anything else.

Do we need user consent before using an AI tool on customer data?

It depends on the lawful ground relied on under the DPDP Act and on the purpose. Consent is one ground, not the only one, and the notice must describe the purposes. Take advice on your specific use.

How often should the policy be updated?

At least once a year, and also when the main DPDP duties begin on 13 May 2027 or when a regulator issues new directions.

This article is general information, not legal advice for any specific case. Lexfolio is a learning resource, and a qualified lawyer should be consulted about your own situation.

Useful books for this topic

As an Amazon Associate, Lexfolio earns from qualifying purchases. Prices and availability are shown on Amazon.

Free CLAT PG 2027 sample: exam guide + 2 judgment passages

Get our 16 page CLAT PG starter free: the 2027 pattern, how to read judgment passages fast, the 25 key Supreme Court judgments of 2026, and two practice passages with 10 explained questions. Plus new free guides and law updates by email. Unsubscribe any time.

Learn this properly, with a certificate. Police and FIRs, consumer and bank fraud complaints, property, wills and inheritance, work rights, family law, cyber fraud and RTI, in plain English. Certificate Course in Everyday Law for Every Indian, ₹349: self-paced lessons, module quizzes and a final test; pass at 60% for a verifiable Lexfolio certificate.

See all Lexfolio certificate courses

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *