LLexfolio

Tag: deepfake

  • Is There an AI Law in India? 8 Myths and the Correct Position

    Last updated: 12 November 2026

    Short answer: India does not have one AI statute. AI use is regulated through laws that already exist: the Digital Personal Data Protection Act, 2023 (DPDP Act), the Information Technology Act, 2000 and its rules, the Consumer Protection Act, 2019, criminal law, and sector rules from regulators such as the RBI. The government’s India AI Governance Guidelines, released by MeitY on 5 November 2025, say a separate AI law is not needed at this stage. That does not mean nothing applies to you. Below are eight beliefs that regularly mislead managers, founders and compliance teams, and the position that actually holds.

    Want a structured walk through AI compliance? 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.

    Myth 1: “India has no AI rules, so nothing applies to us”

    The absence of an AI Act is a drafting choice, not a gap in liability. The Guidelines state that existing laws on information technology, data protection and consumer protection can address many AI risks if they are enforced consistently, while leaving room for targeted amendments in areas such as copyright, data protection and the classification of AI actors and liability. The Secretary of MeitY was reported in the PIB release as saying the focus remains on using existing legislation wherever possible.

    In practice, a loan decision made by a model is still a decision of the lender. A chatbot’s false promise is still a statement of the business. A scraped customer database is still personal data. The law attaches to what the organisation does, not to the tool it uses.

    Myth 2: “The AI Governance Guidelines are binding law”

    They are guidelines, and they lean on voluntary measures such as principles, codes, standards and self-certification. The document itself notes that some voluntary measures may become mandatory baselines as the industry matures. So treat them as a benchmark of reasonable care rather than a statute you can be prosecuted under.

    What they contain is worth knowing, because regulators and courts tend to read new technology through such frameworks. The seven principles (called sutras) are: Trust is the Foundation; People First; Innovation over Restraint; Fairness and Equity; Accountability; Understandable by Design; and Safety, Resilience and Sustainability. The six pillars are Infrastructure, Capacity Building, Policy and Regulation, Risk Mitigation, Accountability and Institutions. Three bodies are proposed: an AI Governance Group, a Technology and Policy Expert Committee and an AI Safety Institute. The Guidelines also recommend that organisations deploying AI offer accessible grievance channels, and that a national database of AI incidents be built, with reporting encouraged without the threat of penalties.

    Myth 3: “The DPDP Act is not in force yet, so personal data in AI tools can wait”

    The DPDP Rules were published on 13 November 2025 and commence in stages. According to the commencement schedule reported by SCC Online, the opening rules (1, 2 and 17 to 21) took effect on 13 November 2025, Rule 4 on consent managers takes effect on 13 November 2026, and Rules 3, 5 to 16, 22 and 23 take effect on 13 May 2027. Rule 3 covers notices, and Rule 10 covers verifiable parental consent for children’s data.

    DateWhat starts
    13 November 2025Rules 1, 2 and 17 to 21
    13 November 2026Rule 4 (consent managers)
    13 May 2027Rules 3, 5 to 16, 22 and 23, which carry most day to day duties of a business

    Waiting is risky for two reasons. First, 13 May 2027 is close, and building a data inventory, vendor terms and a breach process takes months. Second, other duties already bind you today, including confidentiality clauses in client contracts and the general law on negligence and misrepresentation. If staff are pasting client or customer data into public chatbots now, the exposure exists now.

    Myth 4: “If the vendor’s tool makes the mistake, the vendor is liable, not us”

    Under section 8(1) of the DPDP Act, a Data Fiduciary (the organisation that decides why and how personal data is processed) remains responsible for compliance irrespective of any agreement to the contrary, including for processing done on its behalf by a Data Processor. A contract can give you a claim against the vendor, but it does not move your statutory responsibility to the vendor.

    The AI Governance Guidelines point the same way. They recommend graded liability along the AI value chain, proportionate to the function performed, the risk level and the due diligence undertaken. The word that matters is due diligence. An organisation that can show it checked the vendor, limited the data it shared and kept a human check on outputs is in a very different position from one that simply switched a feature on.

    Myth 5: “Anything posted online can be used freely with AI tools”

    Section 3(c)(ii) of the DPDP Act excludes personal data that the person herself has made publicly available, or that someone made public under a legal obligation. The exclusion is narrow. A photograph that a friend posted of you, a leaked database, or a profile compiled from several sources is not automatically outside the Act. Copyright is a separate question: disputes over using copyrighted works to train or prompt models are being litigated in Indian courts and the position has not settled, so avoid assuming that publicly visible means free to ingest.

    Myth 6: “Labelling AI content is only a platform problem”

    The Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules, 2026 were notified on 10 February 2026 and took effect on 20 February 2026. They define synthetically generated information as audio, visual or audio-visual content that is artificially or algorithmically created or altered so that it appears real. Routine edits such as colour correction, compression or noise reduction in good faith are excluded. Permitted synthetic content must be prominently labelled and carry permanent metadata or provenance markers, and platforms may not let users strip those markers. Significant social media intermediaries (those with more than 50 lakh registered users in India) must obtain a user declaration on whether content is synthetic and verify it with technical measures. Orders from the government or a court must be acted on within three hours, down from 36.

    These duties sit on intermediaries, but your marketing team feels them directly: your own AI-assisted posts will be declared, labelled or removed. Separately, the Consumer Protection Act applies to the content itself. Under section 21, the Central Consumer Protection Authority can penalise an endorser of a misleading advertisement up to Rs 10 lakh for a first breach and up to Rs 50 lakh for later ones, and can bar the endorser for up to one year, extendable to three. An endorser who exercised due diligence to verify the claims has a defence.

    Myth 7: “A deepfake is only a reputation issue”

    It can be a criminal one. Section 66D of the IT Act punishes cheating by personation using a computer resource with imprisonment up to three years and a fine up to Rs 1 lakh. Courts have also protected personality rights against AI misuse: the Delhi High Court granted interim protection to actor Abhishek Bachchan in September 2025 against misuse of his name, image and persona, including through AI tools. For an organisation, the lesson is internal: set a rule that staff may not generate the likeness or voice of a real person for marketing without written permission, and keep a takedown route ready for impersonations of your own leaders.

    Myth 8: “Small companies and startups are exempt from data protection”

    There is no blanket exemption for size. The Act lets the Central Government exempt certain classes of Data Fiduciaries, including startups, from some provisions by notification, so check the notification rather than assume. Children’s data deserves special care: section 9 requires verifiable consent of a parent or lawful guardian and bars tracking, behavioural monitoring and targeted advertising aimed at children, subject to exemptions in the Fourth Schedule of the Rules, which cover, for example, educational institutions for educational activities or child safety. Breach of the additional obligations for children can attract a penalty up to Rs 200 crore, and failure to maintain reasonable security safeguards up to Rs 250 crore, as reported when the Rules were notified.

    What to do now

    • List every AI tool in use, including free consumer tools staff use on their own, and note what data goes into each.
    • Write a one page acceptable use rule: no personal data, client secrets or unpublished financials in public tools; approved tools only.
    • Ask each vendor where data is stored, whether inputs train their models, and who handles a breach.
    • Name one person who owns AI risk and one route for staff and customers to report harm.
    • Map the DPDP dates above to a work plan that ends before 13 May 2027.

    Build an AI governance programme with a clear plan. 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 there a law to regulate AI in India?

    There is no single AI law. AI use is governed by existing laws such as the DPDP Act, the IT Act and rules, the Consumer Protection Act and sector regulations. The MeitY Guidelines of 5 November 2025 say a separate AI law is not needed at this stage, though targeted amendments may follow.

    Are the India AI Governance Guidelines mandatory?

    No. They are guidelines that favour voluntary measures. They are still useful as a benchmark of reasonable care, and the document says some voluntary measures may become mandatory baselines later.

    When do the main DPDP duties start for businesses?

    Under the commencement schedule, Rules 3, 5 to 16, 22 and 23 take effect on 13 May 2027. Rule 4 on consent managers takes effect on 13 November 2026.

    Does the DPDP Act cover data that I enter into a chatbot?

    If the data is personal data about an identifiable individual and you decide the purpose of processing, you are a Data Fiduciary, and the Act applies to that processing, including processing done through a vendor on your behalf.

    Do AI generated posts need a label in India?

    The 2026 IT Rules require permitted synthetic content on covered platforms to be labelled and carry provenance markers, and significant social media intermediaries must obtain user declarations. Check each platform’s current tools and your own advertising obligations.

    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.

  • Is It Legal? 6 ChatGPT and AI Situations at Work in India

    Last updated: 12 November 2026

    “Is it legal?” is the question people ask after they have already used the tool. This guide runs through six everyday workplace situations involving ChatGPT and similar tools, and gives for each the legal position in India and the safer way to proceed. Every person and fact below is an invented illustration, written to show how the rules apply; none is a real case.

    Keep in mind that India has no stand-alone AI statute. The India AI Governance Guidelines issued by the Ministry of Electronics and Information Technology on 5 November 2025 prefer voluntary measures and the application of existing laws, so each answer below comes from those existing laws: contracts, the Information Technology Act, 2000 (IT Act), the Bharatiya Nyaya Sanhita, 2023 (BNS), copyright law and the data protection law as it is being phased in.

    Learn this properly. Lexfolio offers AI at Work: Using ChatGPT and AI Tools Safely and Legally, a self-paced certificate course on using ChatGPT and other AI tools safely and legally at work. Course price: Rs 299.

    Illustration 1: The contract pasted into a chatbot

    Facts (illustration): A junior associate at a consulting firm pastes a client’s draft supply agreement, marked confidential, into a public chatbot to “get a quick summary”.

    Legal position. The client’s agreement and the associate’s employment terms almost certainly contain confidentiality clauses, and sending the text to an outside service is a disclosure to a third party. Breach can lead to a contractual claim against the firm and disciplinary action against the employee. Section 72A of the IT Act punishes disclosure of personal information obtained under a lawful contract with intent to cause wrongful loss or gain (up to three years or a fine up to Rs 5 lakh, or both), but intent is needed, so the usual consequences are contractual and professional. If the agreement included personal data of the client’s staff, section 43A and, from 13 May 2027, the data protection law’s safeguards come into play.

    What to do. Tell the supervisor at once. Check the tool’s terms on storage and training use, delete the conversation if the settings allow, and record what happened. If the client’s contract requires notice of a possible leak, give it. In future use an approved tool or summarise by hand with names removed.

    Illustration 2: The market report with statistics that do not exist

    Facts (illustration): An analyst asks a chatbot for “market size data” for a proposal. The answer includes a precise percentage growth figure and a named industry report. The analyst puts both in a client proposal without checking.

    Legal position. Generative tools can produce confident but false numbers and sources. If a client relies on a false statement made carelessly or knowingly in a proposal, the firm can face a claim under the contract or in tort, and a knowingly false statement made to obtain business could raise questions of cheating (section 318 BNS). The Supreme Court’s remarks on 17 February 2026 about a cited case that does not exist, and its order of 27 February 2026 in Gummadi Usha Rani v. Sure Mallikarjuna Rao, show that Indian courts regard unverified output as a failure of professional diligence, not a technical glitch.

    What to do. Withdraw or correct the proposal as soon as the error is found, and tell the client. Trace each figure to a source you can open. Add a rule that statistics and citations in client documents need a named reviewer.

    Illustration 3: The resume screening tool

    Facts (illustration): A recruiter uses a tool that scores resumes and automatically rejects those below a cut-off. A candidate asks why she was rejected and suspects the tool penalised a career break.

    Legal position. There is no Indian statute specifically on automated hiring tools. Private sector employers have wide freedom in hiring, but constitutional equality norms bind state employers and public bodies, and a decision that is arbitrary or based on irrelevant factors invites challenge and reputational harm. The candidate’s personal data in the resume must be handled lawfully: from 13 May 2027 the data protection law will require, among other things, reasonable security safeguards and notice-based processing. The 2025 governance guidelines also encourage deployers of AI to maintain a grievance redressal route.

    What to do. Treat the tool’s score as an input. Let a human review rejections, keep reasons on record and test the tool on past hiring to see whether it screens out a group unfairly. Tell applicants that a tool is used and give a contact for questions.

    Illustration 4: The meeting note taker on a client call

    Facts (illustration): A project manager adds an automatic note taking bot to a video call with a client, without telling the client. The bot stores a transcript on the vendor’s servers.

    Legal position. India has no single statute that sets out in one place when a participant may record a call to which they are a party, so the answer is found in the contract, the client’s confidentiality terms and general privacy and data rules. A transcript may contain personal data and trade secrets, and storing it on a vendor’s servers is a disclosure to a processor. Secretly recording a client can also destroy trust even where no law is broken. From 13 May 2027, personal data in the transcript will fall under the data protection duties.

    What to do. Announce the bot at the start of the call and get a clear yes. Use only an approved vendor, set a retention period, restrict who can see transcripts and delete them on schedule.

    Illustration 5: The advertisement image that looks like a real person

    Facts (illustration): A small brand uses a generated image of a woman who strongly resembles a well known actor to advertise a skin cream, and posts it on social media without a label.

    Legal position. Using a person’s look or voice to suggest endorsement can mislead consumers and may give the person a claim for misuse of personality, passing off or defamation, depending on the facts. The IT Rules were amended by a notification dated 10 February 2026, effective 20 February 2026, requiring permitted synthetic content to be prominently labelled and providing for faster platform action, so an unlabelled synthetic ad risks takedown. Consumer protection law also deals with misleading advertisements. Do not assume that because a picture was “made by a tool” no one’s rights are touched.

    What to do. Use images of real, consenting people with written releases, or of clearly fictional persons, and label synthetic content. Remove the post if the person objects and keep a record of the removal.

    Illustration 6: The voice that sounded like the director

    Facts (illustration): The accounts manager of a mid sized company gets a phone call from a voice that sounds exactly like the managing director, instructing an urgent transfer of Rs 5 lakh to a vendor “for a confidential deal”. The money is sent. It later emerges that the voice was cloned.

    Legal position. Obtaining money by pretending to be someone else is cheating by personation (section 319 BNS) and, when a computer resource is used, section 66D of the IT Act (up to three years and a fine up to Rs 1 lakh). Because the company’s own officer authorised the transfer, the Reserve Bank of India’s 2017 circular on customer liability for unauthorised transactions is unlikely to help much, which makes speed of reporting more important.

    What to do. Call the bank at once, report on cybercrime.gov.in or the helpline 1930 and file a police complaint. A compromised e-mail or account is a listed incident for CERT-In, to be reported within six hours of noticing. Afterwards adopt a call back rule for any payment instruction: verify using a number from your own records, and require two approvers above a set amount.

    What the six situations have in common

    Each problem began with a tool being used faster than the rules around it. The fixes are modest: approved tools, information classes, a named human reviewer, labelling and consent for synthetic content, and a plain way to report incidents.

    Learn this properly. Lexfolio offers AI at Work: Using ChatGPT and AI Tools Safely and Legally, a self-paced certificate course on using ChatGPT and other AI tools safely and legally at work. Course price: Rs 299.

    Frequently asked questions

    Is it a crime to use ChatGPT at work?

    A: No. Using a tool is not an offence. Liability arises from what is done with it, such as disclosing confidential data in breach of contract, cheating or impersonating someone.

    Can a company be sued for something an employee generated with a chatbot?

    A: Yes, in the same way as for any employee’s work, depending on the facts and on whether the employee acted within the scope of the job. A clear policy and training help the employer.

    Do I have to label content made with these tools?

    A: There is no blanket duty on every individual. The amended IT Rules require permitted synthetic content on covered platforms to be labelled, and contracts, clients and institutions may have their own disclosure rules. Labelling is the safe habit.

    What should a company do first if it suspects a leak through an AI tool?

    A: Stop further use, preserve logs and chats, tell the person in charge of security or legal, and check whether the incident is of a type that must be reported to CERT-In or to the client under contract.

    This article is general information, not legal, financial or tax advice for any specific case. Lexfolio is a learning resource and does not provide legal services.