III. Passing Off and Safe Harbour: A problematic Immunity
In the absence of a clearly applicable remedy under the TMA, 1999, the common law remedy of passing off merits consideration. The classic trinity established in Reckitt & Colman v. Borden, 1990, of goodwill, misrepresentation, and damage, fits well in brand impersonation. A synthetic advertisement that exploits an established brand’s goodwill constitutes a misrepresentation, and the resulting damage to the brand is both real and measurable. Indian courts adopted this framework in Laxmikant V. Patel v. Chetanbhagat Shah, 2001 . So, where does the problem lie? It arises from the operation of the intermediary safe harbour under the IT Rules.
Under Section 79 of the IT Act, 2000, an intermediary that observes due diligence under the IT Rules is shielded from third-party liability. Rule 7 of the IT Rules makes it explicit that if the intermediaries comply with the Rules, they get the protection of safe harbour. Now here’s the issue: if an AI generative platform has correctly classified the brand impersonation deepfake as falling under Proviso (b) of Rule 2(wa), it has no SGI-related due diligence obligation in respect of that content. It has therefore complied with the Rules and gets the safe harbour against a passing off action.
Notwithstanding this qualification, a structural difficulty remains in the AI brand impersonation. It is necessary to distinguish here between two categories of platform: the AI generative platform that produces the synthetic content and the social media intermediary on which content is subsequently hosted. Both may, in principle, invoke the safe harbour under Section 79 read with Rule 7 of the IT Rules, but the analysis differs in important respects, and the distinction has direct consequences for the remedial options available to a trademark owner.
Where the AI generative platform has correctly classified the brand impersonating content as falling within Proviso (b) of Rule 2(wa) and accordingly as not constituting SGI as it bears no SGI-specific due diligence obligation in respect of that content. Having complied with the applicable Rules, it qualifies for safe harbour protection and may resist a passing off claim on that basis, unless and until actual knowledge of the specific infringement is established. This results in creating a double shield where the AI platform may contend that it was under no legal obligation to prevent publication of the brand impersonating content and it bears no civil liability for having failed to do so.
It is essential to understand the precise nature of this immunity that it is conditional rather than absolute. Section 79(3) provides that the safe harbour ceases to apply once the intermediary has ‘actual knowledge’ of the unlawful character of hosted content and fails to expeditiously remove it. Upon receipt of a properly constituted notice, therefore the platform acquires actual knowledge and an obligation to investigate and, where warranted, remove the content arises. Thus, a safe harbour operates as a conditional pre-notice immunity and not a permanent and absolute shield against liability.
Furthermore, the actual knowledge standard in introduces a further complication specific to the AI generative platform as distinct from the social media intermediary. A social media intermediary that hosts user-uploaded content may plausibly claim that it had no knowledge of the content prior to notification. However, an AI generative platform is the entity whose model produced the infringing output and it is difficult to sustain the argument that a platform has no knowledge of content that its own systems generated and response to a user prompt. Whether the generation of infringing output by an AI platform’s model itself constitutes actual knowledge, such that the pre-notice immunity period is eliminated entirely for that category of platform, is a question that the Indian judiciary has not yet resolved, but it is a question that goes directly to whether section 79, as currently drafted, is coherent in its application to AI generative platforms at all.
IV. A typology of AI Brand impersonation: Three categories, Three different Issues
AI-generated brand impersonation is not legally homogeneous. It differs materially depending on the nature of the intellectual property implicated, the identity of parties involved in the generation and dissemination chain, and the specific provisions of the IT Rules and the TMA 1999 that are involved. Such impersonation falls into approximately three categories:
First, Logo & Trade Dress where a synthetic ad replicates your registered logo and packaging that shows no recognisable person. Rule 2(wa) is inapplicable in this case as it requires depiction of an individual or event. A registered logo is not an individual. The synthetic reproduction of a visual brand identity is not an ‘event’ in any legal sense or ordinary sense of that term. The content therefore falls outside the SGI framework under The IT Rules, and its labelling obligations, metadata requirements, monitoring duties and takedown windows have no application. Under the TMA, 1999, while logo and trade dress protection exists in principle for registered marks, the ‘use in the course of trade’ attribution question remains unresolved. As argued in Part II, Section 29 requires ‘use in the course of trade’ by an identifiable defendant. In the logo impersonation scenario, the user who prompted the generation may be unidentifiable. The AI platform generated the output but, as argued above, the platform’s generative act does not correspond to conventional trademark ‘use’ in the sense of affixing a mark to goods or deploying it as a commercial identifier. There is a registered right; however, no effective enforcement mechanisms against such AI infringers are presently available.
Second, Synthetic Celebrity Spokesperson, where an AI-generated video uses your ambassador’s voice and likeness. Rule 2(wa) technically applies insofar as an identifiable individual is depicted, but Proviso (b)’s “hypothetical/research” is available to the platform. Proviso(b)’s broadly framed exemptions remain available to the AI platform on a self-assessment basis and may operate to make the SGI framework inapplicable. A generative AI platform that describes the synthetic celebrity endorsement as a ‘demonstrative output for content moderation training may invoke Proviso (b) without any ex-ante or independent scrutiny, thereby disapplying the SGI due diligence framework entirely before the brand owner or the ambassador can engage the grievance mechanism. Even where the SGI under IT Rules applies, the passing-off action based on misrepresentation faces the distributed attribution problem as identified in Part II. Thus, the ambassador’s contractual image rights protect the ambassador in relation to the brand owner, but do not directly confer on the brand owner an independent cause of action against the AI platform or social media intermediary responsible for the synthetic depiction.
Third, an AI-generated Sound Mark where a synthetic audio clip reproduces your registered sound mark , a distinctive jingle or branded audio identity with sufficient fidelity to cause confusion as to trade origin. Rule 2(wa) is inapplicable as neither an individual nor an event is depicted, and the concept of an ‘event’ does not naturally encompass the reproduction of a registered audio mark. There is no ambiguity here, no case to be resolved by purposive construction, as the intention of the legislature is very well clear in the definition that cannot bring the sonic dimension within the provision.
The TMA 1999 protects registered sound marks under Section 2(1)(zb), read with Rule 26(5) of the Trade Marks Rules, 2017, but again, whether it recognises AI synthesis of a registered sound mark is not provided, even where it amounts to brand impersonation and perceived confusion among consumers, the answer is that current law does not clearly provide a remedy. The question is not merely one of attribution but also one of whether the AI-generated audio output, which is a statistical approximation of acoustic patterns derived from training data rather than a direct copy of the registered mark, satisfies the ‘identical or deceptively similar’ standard under Section 29. Thus, IT Rules offer no remedy while TMA, 1999, furnishes a registered right but provides no mechanism to test its vindication against an AI mediated infringer.
Across all three categories, the pattern is consistent. The registered IP right may be technically subsisting, the commercial harm may be demonstrable and quantifiable and yet the available remedies are either structurally inapplicable or procedurally invalid. Thus, the brand owner is left to a registered right that the law has technically safeguarded but in practice cannot effectively implement against infringement by AI in all three types of actions. However, the legal remedy is either non-existent or even lacked by the IT Rules, 2026, that was supposed to help.
V. What Needs to be Changed
There are certain addressable and specific deficiencies in the existing legal framework, which can be addressed in a targeted and coherent manner.
Firstly, the “educational and research output” exemption under Rule 2(wa) should be made subject to a mandatory floor. If the content incorporates a registered trademark, a recognizable celebrity likeness with a commercial endorsement relationship or a registered sound mark, it cannot be exempted, regardless of the stated purpose. The question of “use in the course of trade” as provided in Section 29 of the TMA cannot go on indefinitely. This would not eliminate the educational and research exemption in its general application but would establish a minimum threshold below which platform self-characterisation cannot operate to disapply the SGI framework in cases of evident commercial harm. The reform is capable of implementation by amendment to the proviso without requiring structural revision of the IT Rules as a whole. In addition, the determination of whether content falls within Proviso (b) should, in cases involving registered intellectual property, be subject to review by an independent regulatory body rather than left to platform self-assessment.
Further, Generative AI platforms that are registered as Significant social media intermediaries under Rule 2(w) of the IT Rules should be required, as a condition of retaining safe harbour protection, to implement automated trademark screening mechanisms at the point of content generation. Such mechanisms, analogous to the Content ID systems already deployed by major video hosting platforms for copyright compliance, would flag outputs incorporating registered marks and require either platform-level review or user-level disclosure of commercial purpose before the content is made available for download or sharing.
Secondly, the ‘use in the course of trade’ requirement under Section 29 of the TMA 1999 requires legislative explanation in the AI context. The legislature should expressly provide that the generation, by an AI platform’s model, of output that incorporates or closely replicates a registered trademark constitutes use of that mark in the course of trade for the purposes of Section 29, regardless of whether the act of generation is the direct act of a natural person. Absent legislative action, Indian courts should be invited, in an appropriate case, to extrapolate the reasoning of Satyam Infoway to AI-generated infringement.
Thirdly, there is a compelling case for regulatory coordination between MeitY (Ministry of Electronics and Information Technology) and the Office of the CGPDTM. These two bodies presently operate in entirely separate regulatory frameworks with no coordinated jurisdiction over the intersection of AI generated content and IP rights. It is proposed that MeitY and the CGPDTM establish a shared notification protocol under which a brand owner that files a trademark infringement complaint with the CGPDTM simultaneously triggers a corresponding takedown obligation under the IT Rules framework, without requiring the filing of a separate grievance with the platform. A memorandum of understanding between two bodies, establishing a unified grievance channel for AI-generated brand impersonation complaints, a common evidentiary standard for takedown requests, and a periodic inter-agency review mechanism, would materially reduce the procedural burden on trademark owners.
Fourthly, the law should establish a clear respondent hierarchy for AI-generated trademark infringement. Under the existing framework, a brand owner confronting AI-generated trademark infringement faces genuine uncertainty as to which party bears primary legal responsibility, whether its AI platform, the individual user or the social media intermediary, with liability capable of being deflected between all three without resolution. At least, the law ought to establish the AI platform as the initial respondent in case of content generation by its model, and allow it to determine the contribution thereof. It is proposed that an amendment to Rule 3 of the IT Rules, which already governs the due diligence obligations of intermediaries, expressly designate the AI generative platform whose model produced the infringing output as the primary respondent in any complaint or proceeding arising from that output. It will require an AI platform to respond to the complaint, investigate, and either remove the content or disclose the identity of the user to the owner. The AI platform would retain the right to seek contributions from the individual user or the social media intermediary in subsequent proceedings. This would relieve the brand owners of the burden of pursuing three separate parties to obtain removal of a single infringing advertisement.
Lastly, Section 79 of the IT Act, 2000 requires reformulation to reflect the qualitative distinction between platforms that passively host third-party content and platforms that actively generate content in response to user prompts. Section 79 was designed for platforms that contain content and not for platforms that create it. It does not provide for a distinction between YouTube serving an unrealistic ad of a user and an AI-based tool that creates such an advertisement on demand. An AI generative platform that produces synthetic output incorporating a registered trademark on command has engaged in an act qualitatively different from a video hosting service that stores a user-uploaded file. The Rules ought to make that line very clear that safe harbour does not extend to apply to AI-generated output where the platform itself has used the registered mark.
VI. The Bigger Picture
Returning to the opening scenario presented in the introduction: What does the Indian law currently offer to resolve the problem of a synthetic video of AI-generated content incorporating a registered logo, a trademarked jingle and a voice indistinguishable from the company’s contracted ambassador that has been online for 12 hours and has eight hundred thousand views?
If the video does not depict a recognisable face, it falls outside the IT Rules entirely, prior to any consideration of exemptions. If it does depict the ambassador’s likeness, the AI generative platform may invoke the Proviso (b) exemption through self-characterisation as demonstrative or educational content. The TMA 1999 provides no authoritative judicial guidance on which party in the AI generation chain constitutes an infringer for the purposes of sec 29. The passing off action is available in principle, but the Sec 79 safe harbour may insulate the relevant platform from civil liability unless and until actual knowledge of the specific infringement is established through a properly constituted notice. A John Doe order may secure interim relief against unidentified individual users, but does not resolve the underlying question of platform liability.
The 2026 amendment to India’s IT Rules, by the comparative standards of global synthetic media regulation, is a considered and substantive piece of drafting. The labelling mandate, the three-hour take-down window, and the metadata and unique identifier requirements deserve meaningful interventions in the governance of synthetic media. The problem isn’t ambition but the amendment was drafted as a response to the threat of non-consensual deepfake content targeting individual harm. Brand harm, the commercial, industrial-scale, impersonation of logos, sound marks, and visual identities, was simply not on the drafting table.
Meanwhile, the Trade Marks Act 1999 has not yet been stretched by judicial authority to cover a generative AI model that produces a fake brand advertisement in four seconds on a free-tier subscription. A brand owner in India today has fewer clear legal options against an AI-generated brand impersonation campaign than they do against a street vendor selling counterfeit goods. That is not a tolerable state of affairs for any serious IP jurisdiction, and it is certainly not consistent with India’s ambition to be a leading digital economy.
(This post has been authored by Priyanshi Jain, 4th Year student at National Law University, Jabalpur)
CITE AS: Priyanshi Jain, ‘AI Generated Brand Impersonation and the Limits of ‘Use in Course of Trade’ under the Trade Marks Act, 1999’ (The Contemporary Law Forum, 06 July 2026) <https://tclf.in/2026/07/06/ai-generated-brand-impersonation-and-the-limits-of-use-in-course-of-trade-under-the-trade-marks-act-1999-part-ii/> date of access.