MediaNama attended a closed-door discussion under Chatham House Rules on how e-commerce, artificial intelligence and illicit trade are converging in India. This was a group consisting of brand owners, law firms, customs and enforcement officials, intergovernmental agencies, foreign intellectual property offices and researchers. No e-commerce platform was in the room, and this led to some interesting perspectives.
Why it matters: Across three hours, the recurring complaint was that platforms hold the data, set the commercial incentives and control the onboarding – and that none of those is currently answerable to anyone outside the platform. That case was made in full, and unopposed.
The KYC problem, described from the outside
Seller verification was the thread that ran through the afternoon, and the criticism was structural rather than anecdotal.
Speakers described know-your-customer (KYC) checks as the foundation of platform trust, and as the part of it that does not hold. Registered seller addresses, several said, turn out on inspection to be empty plots or to not exist.There was broad support for movingfrom a one-time check at onboarding to continuous, risk-weighted verification.This should be appliedmore heavily tosellers with a history of counterfeit complaints,opined the speakers.
A related practice was described in some detail: networks of intermediaries that exist to get flagged sellers back onto platforms, using fabricated brand authorisation letters, forged invoices and, in some cases, counterfeit government seals and logos.The proposed answer was standardised onboarding checksacross e-commerce platforms,andshared alerts.This is to ensure thata seller removed from one marketplace cannot simply reappear on another.
The incentive nobody in the room could fix
One of the most important points of the afternoon was about money.
An e-commerce platform earns its commission on a sale whether the goods are genuine or counterfeit. And a seller with confirmed infringement complaints against it can, on the account given in the room, still buy advertising placement, carry a recommended tag, or win the buy box. Nobody could identify a policy at any platform tying enforcement outcomes to a seller’s commercial privileges.
The data point followed the same shape. Information about bad actors exists, but it is fragmented across brand owners, platforms and agencies. The speakers alleged there is little or no appetite among e-commerce platforms for exchanging it voluntarily. One observation was that platform willingness to share improveswhen a platform’s market position comes under competitive pressure, and not otherwise.
Where intermediary liability is being stretched
A second thread questioned whether the distinction betweenan active and a passive intermediarystill means anything.
The argument put was that platforms now run recommendation engines, curate listings, operate fulfilment and warehousing. They also sell their own private-label brands alongside third-party sellers and mixed inventory. A platform that actively promotes a listing, including a potentially infringing one, is not obviously passive in respect of it. Regulators elsewhere are already moving to hold platforms accountable for automatically promoted content, and the same logic was said to extend to e-commerce.
A newer risk was raised alongside it. AI shopping agents acting on a consumer’s behalf are not currently equipped to tell a genuine listing from a counterfeit one, and could be manipulated into buying the wrong thing.
On AI and copyright, three unresolved problems
The second session moved to intellectual property and AI, and three problems came up that Indian law has not answered.
The first was whether the reasoningin India’s recent AI-copyright ruling holds together. A court has accepted that training may qualify as fair dealing but appeared to tie that conclusion to the absence of infringing output. Ingestion and output are, in principle, separate legal questions, and several speakers thought the linkage deserved closer examination.
The second wasremedies. Once infringing material is absorbed into a model, it generally cannot be selectively removed the way a file can be taken down from a website. A court cannot practically order extraction. That, it was suggested, may push Indian courts toward damages rather than injunctions.
The third was access to licensing. Voluntary content licensing deals are being struck between the largest rights holders and the largest AI developers. Small and mid-sized rights holders have no comparable route, and no mechanism was identified that would give them one.
There was also a procedural observation worth noting for anyone following Indian injunction practice. Dynamic-plus injunctions, which allow a rights holder to act against future and still-unidentified infringing domains without returning to court each time, were described as a meaningful development. They were presented as a solution. They are also, from the other direction, an order against people who have not yet been identified.
Every criticism above is serious, and none of it was answered, because there was nobody present to answer it.
The organisers’ account of the design is that platforms were left out of this first session in order to consolidate industry and government concerns first, with a second round including platforms planned for later in the year. That is a coherent way to run a consultation. It is not a basis for policy, and the organisers appear to accept as much. They convened that the second round is among the stated next steps, alongside circulating a policy summary to the Controller General’s office and to the Department for Promotion of Industry and Internal Trade, and setting up a working group of brand owners to pilot a shared format for seller and KYC data.
Whether any of that produces a rule is a separate question. None of it binds anyone, and India’s e-commerce enforcement framework is at present being shaped case by case through litigation rather than through legislation.
- 8 streaming services, 35 content complaints, 7 upheld. None checking reports
- Microsoft agreed not to train AI on US student data. India has no such legal bar until May 2027
- Balaji forecast revenue for seven business lines, down to its astrology app. Its AI pillar got no number
