Generative AI has moved from a novelty in advertising to the default production layer for a large share of the world’s marketing content — writing scripts, generating “actors,” cloning voices, and fabricating entire scenes that never happened. The advertising world has not caught up equally everywhere. India is currently governed by a patchwork of self-regulatory guidance and freshly amended IT rules, while the European Union and, to a lesser extent, the United States, are moving toward binding, enforceable disclosure regimes with real financial teeth.
Why Generative AI in Advertising Needed Guardrails in the First Place
The appeal of generative AI to advertisers is obvious: a brand can produce a synthetic spokesperson, generate hundreds of localized video variants, or simulate a product demonstration without a shoot, a studio, or a single actor’s contract. But the same capabilities that cut cost and turnaround time also make it trivially easy to fabricate an endorsement, exaggerate a product’s performance, invent a location, or clone a real person’s face and voice without consent.
Regulators across jurisdictions have converged on the same worry: when a consumer cannot tell whether an image, a doctor’s opinion, or a customer testimonial is real or synthetic, the basic premise of truthful advertising collapses. The question every framework is now trying to answer is not whether generative AI should be allowed in advertising — nobody is proposing a ban — but when and how its use must be disclosed, and where the line sits between an unregulated creative tool and a deceptive practice.
India’s Regulatory Landscape: From Soft Guidance to a Late-Stage Draft Framework
The Early Years: A White Paper, Not a Rulebook
India’s engagement with generative AI in advertising began not with legislation but with an advisory document. In August 2023, the Advertising Standards Council of India (ASCI), working with law firm Khaitan & Co, published a white paper on the opportunities, risks, and legal considerations of generative AI in advertising. It flagged real gaps: AI is not recognized as a legal entity under Indian copyright law, meaning advertisers may struggle to claim ownership over AI-generated creative or to pursue infringement claims against copycats. It also warned that generative models are frequently trained on personal data scraped from the public domain, and that India’s absence of a dedicated data protection law at the time left advertisers exposed when deploying AI chatbots or personalization tools that might ingest sensitive consumer information.
Crucially, this document had no enforcement mechanism. It was guidance for advertisers to self-audit against, not a code they could be penalized for violating.
The IT Rules Amendment and the Arrival of “Synthetically Generated Information”
The more consequential shift came through India’s information technology framework rather than advertising law specifically. The Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules, amended in February 2026, introduced the concept of “Synthetically Generated Information” (SGI) — content that results from AI technologies that alter or produce new material to create false representations of actual events or persons. The amendment requires visible labelling of qualifying synthetic content, including a watermark covering a meaningful share of an image or video’s surface area, and it gives platforms an obligation to act on flagged deepfakes within a short window.
This was the first time Indian law reached directly into the mechanics of AI content labelling, rather than leaving the matter to industry self-regulation. It also gave ASCI a legal peg to hang a more formal advertising-specific framework on.
ASCI’s Draft Guidelines: A Risk-Tiered Approach, Still Only a Draft
On 8 May 2026, ASCI published its Draft Guidelines for Responsible Labelling of AI-Generated Content in Advertising, explicitly aligned with the amended IT Rules. Rather than mandating a blanket disclosure on every advertisement touched by AI, the draft adopts a three-tier, risk-based model:
- High risk (prohibited outright, label or no label): fabricated endorsements or testimonials, AI-generated “doctors” or authority figures implying medical credibility, exaggerated product claims conveyed through AI-generated visuals, fabricated locations presented as real, and deepfakes or likeness replication used without consent. These are barred regardless of whether a disclosure label is present.
- Medium risk (mandatory disclosure): synthetic influencers or brand ambassadors, AI-generated product demonstrations, AI-recreated environments or sound effects, and AI-driven sponsored product recommendations. ASCI has proposed sample labels such as “Audio/Video created using AI” or “Audio/Video enhanced using AI,” while leaving brands flexibility to use any wording that clearly informs the consumer. Even where a celebrity has consented to a synthetic recreation of their voice or likeness, disclosure is still required — consent does not remove the labelling obligation.
- Low risk (no labelling required): routine colour correction, blemish removal, minor lighting adjustment, decorative or ambient AI-generated backgrounds and music, and administrative uses such as AI-assisted copywriting or accessibility descriptions.
The draft was open for public consultation until 13 June 2026, and as of the time of writing it remains a draft rather than a binding code — advisory in tone, dependent on ASCI’s self-regulatory enforcement powers (which extend to referring non-compliant advertisers to sectoral regulators, but not to imposing fines directly).
How the European Union Has Approached the Same Problem
The EU’s answer sits inside the broader Artificial Intelligence Act rather than in a standalone advertising code, but its Article 50 transparency obligations reach deep into commercial content. Providers of generative AI systems must ensure that outputs — audio, image, video, or text — are marked in a machine-readable format and detectable as artificially generated. Deployers using AI to create a “deepfake,” defined as synthetic image, audio, or video content that resembles a real person, object, place, or event closely enough to appear authentic, must disclose that fact to viewers, with only narrow carve-outs for evidently artistic, satirical, or fictional work.
Two features distinguish the EU’s approach from India’s current draft. First, it is binding law, not a self-regulatory code: penalties for non-compliance can reach €15 million or 3% of a company’s total worldwide annual turnover, whichever is higher. Second, the European Commission has been explicit that weak disclosure — a mention buried in terms and conditions, a watermark alone, or a vague label like “assistant” — does not satisfy the requirement. Guidance under development calls for clearly visible plain-language notices at the point a consumer first encounters the content, not passive technical metadata. A voluntary Code of Practice, developed alongside the binding rule, is expected to give companies a de facto compliance template even before enforcement guidance is finalized. These obligations become applicable from 2 August 2026, with some marking-and-detection elements potentially receiving a short transitional grace period.
The United States: Agency Guidance and an Emerging Patchwork of State Laws
The US has no federal AI-advertising statute comparable to the EU’s AI Act. Instead, the Federal Trade Commission has relied on its existing authority over deceptive practices, issuing staff guidance in 2025 built around three principles: AI involvement in generating or substantially modifying advertising content should be disclosed to consumers; claims made through AI-generated content must be truthful and substantiated just as human-made claims are; and AI-generated endorsements or testimonials that create a false impression of a real person’s experience are treated as deceptive regardless of whether a disclosure is present. In parallel, individual states have begun legislating directly — New York, for instance, passed a dedicated AI content disclosure law for advertising, signed in January 2026, requiring clear disclosure when content is “substantially generated” by AI.
The result in the US is a hybrid: strong enforcement authority (per-violation penalties that can scale into the millions of dollars across a large campcampaign) sitting on top of guidance rather than a single comprehensive statute, with individual states beginning to fill gaps unevenly.
Comparing the Three Regimes
| Dimension | India | European Union | United States |
|---|---|---|---|
| Legal status | IT Rules amendment (binding) + ASCI code (draft, self-regulatory) | AI Act Article 50 (binding statute) | FTC guidance + patchwork state laws |
| Structure | Three-tier risk classification (high/medium/low) | Deepfake and synthetic-content disclosure obligations across the value chain | Principles-based; disclosure tied to deception standard |
| Enforcement power | ASCI referrals to sectoral regulators; no independent fines | Fines up to €15 million or 3% of global turnover | Per-violation FTC penalties; state-level penalties vary |
| Consent exception | Consent does not remove disclosure duty | Consent does not remove disclosure duty | Consent plus disclosure both required for voice cloning |
| Maturity | Draft stage, consultation closed June 2026 | Applicable from August 2026, guidelines still being finalized | Guidance issued 2025; state laws emerging in 2026 |
Where India’s Gap Actually Lies
The comparison suggests the gap is not conceptual — India’s risk-tiered logic is, if anything, more granular than the EU’s binary deepfake test, and its examples (a fabricated “doctor” endorsing a supplement, a fake location, an undisclosed synthetic brand ambassador) map closely onto the same harms the EU and US are targeting. The gap is institutional and procedural:
- Binding force: The ASCI framework, however well designed, remains a self-regulatory code. Historically, ASCI’s sanctions extend to publicizing non-compliance and referring persistent violators to bodies like the Ministry of Information and Broadcasting or sector regulators — a materially softer deterrent than statutory fines calculated as a percentage of global turnover.
- Timing: India’s draft guidelines are still in consultation, roughly contemporaneous with the EU’s binding rules coming into force. A regime that is still being finalized when peer economies are already enforcing theirs risks a compliance lag for multinational advertisers running the same creative across markets.
- Copyright and ownership ambiguity: ASCI’s own 2023 white paper flagged that AI-generated advertising creative may fall outside copyright protection in India because AI is not a recognized legal author. This unresolved question sits alongside the newer disclosure rules and has not been directly addressed by the 2026 draft.
- Data protection interlock: India’s Digital Personal Data Protection Act adds a privacy layer that the advertising-specific guidelines do not fully integrate, particularly around what happens to consumer data fed into AI personalization or chatbot tools during a campaign.
What This Means for Advertisers and Agencies Operating in India
For marketing teams and agencies working across Indian and international campaigns, the practical takeaway is to plan for the stricter regime rather than the most lenient one. Since ASCI’s proposed high-risk category already prohibits fabricated endorsements, undisclosed deepfakes, and synthetic authority figures outright — irrespective of any label — and since the IT Rules amendment already carries statutory watermarking requirements for synthetic content, the safest posture is to treat disclosure as the default for any AI-assisted asset that could plausibly influence a purchase decision, rather than waiting for the ASCI draft to be finalized. Brands running campaigns simultaneously in India, the EU, and the US will find that building to the EU’s stricter, visible, plain-language disclosure standard largely satisfies India’s medium-risk labelling expectations as well, while the reverse is not necessarily true.
The direction of travel across all three jurisdictions is unmistakable even where the instruments differ: consent from a person whose likeness or voice is used no longer substitutes for disclosure to the audience, technical metadata alone is treated as insufficient, and the harm regulators care most about is the same everywhere — a consumer relying on something that looks real but isn’t. India’s framework is converging toward that same standard; what it currently lacks, relative to the EU in particular, is the binding force and finalized timeline to back it up.
