Licensing and Labelling of AI-generated content in India

Generative Artificial Intelligence (“AI”) has greatly transformed content creation. Tools such as ChatGPT and DALL-E have enabled the rapid production of text, images, audio, and video. However, despite its merits, generative AI has also raised serious challenges around misinformation, accountability, bias, and copyright. Surveys show that 63% of users face inaccuracies in AI-generated content, and 94% of consumers want mandatory disclosure of such generation.

In view of these circumstances, the Standing Committee on Communications and Information Technology, headed by MP Nishikant Dubey, has recently proposed new rules for licensing and labelling of AI-generated content. The draft recommendations call for mandatory labelling of AI-generated media, licensing for content creators, and joint oversight by the Ministry of Information and Broadcasting and the Ministry of Electronics and Information Technology (“MeitY”) to establish techno-legal measures for identification and prosecution of abuse, and institutional arrangements such as a media fact-checking mechanism. The proposed licensing mechanism will also bring accountability and ensure that a creator can be held responsible if their work causes harm. While these committee reports are not legally binding, they often form the basis of legislative reforms in India. MeitY has also been working on a similar law that would require social media platforms to add watermarks and labels to AI-generated content.

India currently lacks a dedicated law on regulating AI, and this urgency to regulate it is primarily due to AI’s potential to amplify fake news and disrupt public order. Similar concerns were also raised in the Karnataka Misinformation and Fake News (Prohibition) Bill, 2025, which was considered overly broad but showcased the threat of digital misinformation to society. With the use of AI growing rapidly, and the Indian market estimated to reach US $17 billion by 2027, the need for clear regulations has become even more pressing.

In this backdrop, this piece aims to critically examine India’s emerging proposals on licensing and labelling AI-generated content. It highlights constitutional and practical challenges, lessons from the European Union (“EU”), United States (“US”), and China, and finally proposes a balanced framework for the regulation of AI-generated content.

Challenges and considerations for India’s proposal
The proposed regime, which combines licensing and labelling of AI-generated content, while well-intentioned, suffers from certain shortcomings. Licensing, in particular, would be largely unprecedented globally. No major jurisdiction, not even China with its notoriously strict AI controls, requires individual creators to obtain a license before generating content. Historically, India has suffered from the inefficiencies of a “license raj”, and licensing requirements for speech and expression have been scrutinised as a form of prior restraint and been held to be permissible only under the narrowly tailored exceptions in Article 19(2). Further, it also raises certain important constitutional questions. If creators are required to obtain licenses, what happens when one is revoked? Will there be an appeal or will administrative discretion prevail? Imposing such control over content creators has the risk of bureaucratic overreach and a chilling effect on free speech.

Mandatory labelling may be less intrusive, but certain questions still remain. What is considered “AI-generated”? Would a journalist who drafts articles with ChatGPT but edits them manually also be required to label the article? Would an artist who simply uses AI filters for improvement be treated the same as someone using a deepfake created by a bot? This is important because labelling can even unfairly stigmatise works that are only lightly edited by AI, as public perception often equates an “AI tag” with lower credibility.

Mass content monitoring may also tempt platforms to adopt automated monitoring systems, which could lead to an over-regulation of digital speech. What is important here is to balance platform responsibility and creator responsibility, i.e., intermediaries’ safe harbour protections under the IT Act must be retained, but content creators should not have a disproportionate compliance burden on them. So, while labelling has its merits, it also has certain glaring flaws; thus, to navigate them effectively, India can benefit from examining how other countries regulate AI-content labelling, as will be discussed in the following sections.

Global comparative perspectives on AI-content regulation
The EU AI Act, coming into force in August 2026, sets out detailed transparency obligations under Article 50. Providers must disclose when individuals are interacting with AI systems, unless it is already obvious. Further, all artificially generated outputs, such as text, audio, video or images, must be machine-readable and detectable as artificially generated or manipulated. However, under Article 50(2), this obligation shall not apply to AI systems that do not substantially alter the input data, such as for standard editing or grammar fixes, etc. The Act also mandates clear disclosure for deepfakes and public-facing AI text. Although artistic, satirical or editorially reviewed content is exempt. The Act also creates an AI Office for uniform monitoring and facilitation of compliance with these transparency obligations across the EU.

In contrast, the United States takes a sector-specific and enforcement-based approach. The proposed AI ​​Labelling Act of 2023 (H.R. 6466) mandates that AI-generated text, audio, and visual content contain clear and conspicuous disclosures. The metadata must record the tools used, timing, and nature of AI involvement, and place contractual obligations on developers and licensees to ensure that they are not deleted even at a downstream level. Violations fall under the Federal Trade Commission’s “unfair or deceptive practices” clause, which carries civil penalties. Complementing federal efforts, California’s AI Transparency Act (SB 942, 2024), set to become operative on January 1, 2026, requires large generative AI providers to embed visible and hidden disclosures in AI-generated content, offer free detection tools, and impose $5,000 per-violation penalties.

However, China has taken the most stringent approach and has incorporated AI-labelling into its information-control regime. Under the Measures for the Administration of AI-Generated Content Labelling and the mandatory national standard GB 45438-2025: Cybersecurity Technology — Labelling Method for Content Generated by Artificial Intelligence, regulations mandate both explicit and implicit disclosure, with metadata added at the point of creation and retained during distribution. Platforms bear the primary enforcement responsibility and are obligated to block unlabeled content and report violations to the authorities. The regulations apply not just to deepfakes and generative media but also to text, and impose strict liability for non-compliance. This strict enforcement model is precisely the stringent, state-dominated model that we fear in India and should prefer not to emulate.

The way forward: Lessons from global regimes
While India’s licensing plan for AI content creators will make it one of the first countries to enter this uncharted territory, China and the United States illustrate two distinct strategies for regulating AI-generated content. China’s measures require artificial content to have built-in metadata and make platforms directly liable for monitoring and removal. While this is effective for traceability, it risks overregulation. And legitimate uses, such as satire or political criticism, could also be stifled, which would go against India’s constitutional guarantees of Article 19(1)(a). In the US, on the other hand, the AI Labelling Act of 2023 remains stalled in Congress, with both the Senate and House bills having seen no progress beyond committee referral, while the Californian AI Transparency Act carries only a limited territorial mandate. While these measures are promising in intent, their effectiveness remains untested, making it difficult to assess their potency. It is this uncertain and uneven compliance landscape that makes the American model an unreliable guide. India must therefore avoid both extremes: China’s state-centric censorship model and the US’s inefficient, patchwork regime.

The EU AI Act, on the other hand, presents a comparatively more balanced framework by avoiding both the rigidity of China’s blanket mandates and the uncertainty of the United States’ fragmented, untested regime. Firstly, Article 50 sets out different transparency obligations for different AI systems: disclosure when interacting with conversational AI [50(1)], labelling of synthetic text, audio, image and video outputs [50(2)], notification for emotion recognition and biometric categorisation systems [50(3)], and disclosure of deepfakes and public-facing AI-generated text [50(4)]. Secondly, Article 50(2) requires that AI outputs be marked in a machine-readable format and be detectable as such, and also exempts cases where AI is used only for standard or assistive editing that does not substantially alter the user’s input. Thirdly, Article 50(4) then strikes a balance with human oversight, as when AI-assisted content undergoes human review, the labelling requirement is lifted.

Firstly, for India, this means that licensing or disclosure should be limited only to high-risk outputs, such as deepfakes or content that could influence elections and public order, as was similarly mandated by the Election Commission, and not to simple memes or artistic works. Secondly, India should adopt metadata standards for AI outputs, enabling machine detection and visible labelling for human users, and exclude standard AI-assisted edits that do not substantially alter the input. Thirdly, India can reduce the burden on news organisations and independent creators by adopting human review exemptions, most importantly, by issuing practical guidelines on what constitutes human oversight and editorial review. More specifically for India, it should narrowly define high-risk AI content (such as deepfakes in elections, health, or financial misinformation) to avoid overreach, adopt the licensing model in phases and simplify it for small creators, implement tiered penalties instead of blanket fines, and establish an AI regulatory committee, like the EU’s AI Office, to monitor compliance with the above. NUJS Law