Should AI Companies Be Charged For Using Content To Train Models?

Artificial intelligence tools learn from huge volumes of media shared online. News articles, books, songs and artwork all feed into how these systems learn language and style. In most cases, creators never receive payment when their work gets used in this way.

India now wants to change that. A panel set up by the Department for Promotion of Industry and Internal Trade announced a proposal that would link AI training to payments for copyright holders. The idea centres on payment rather than permission.

The panel’s view is that creators should receive money when commercial AI tools use their work. At the same time, it does not want to slow AI development inside the country.

What Does The Proposed Licence System Look Like?

The proposal sets out a blanket licence system. Under this setup, AI companies would train models on lawfully accessed Indian content without contacting each creator directly. Books, films, music, news stories, and artwork all fall under this system.

Instead of many private deals, AI firms would make one payment to a single national body. That body would collect royalties and pass the money on to rights holders. Independent writers and artists would sit alongside large publishers and labels.

A government-appointed committee would decide royalty rates. Payment would apply only once an AI product enters commercial use. Tools built for hobby use or free testing would sit outside the payment rule.

The panel described this as a one national licence, one payment system. It aims to lower paperwork for AI builders while making sure creators receive money for the use of their work.

What Could This Mean For AI Firms, Creators and Users?

For AI developers, the proposal may raise costs once products reach the market. Training large models already costs millions of dollars, and royalty fees could add another expense. These costs may pass through to users over time.

For creators in India, the system could open a new income stream. Writers, musicians, artists, and journalists could receive money when AI tools train on their published work. Payments would arrive through the central body rather than direct deals.

Users of AI tools may also notice changes. The proposal points toward better records around training data and stronger respect for copyright rules. This could lead to more transparency around how AI systems learn.

India’s proposal does not stop AI development. It redraws the rules around ownership and payment. The government will gather feedback on the proposal over the coming weeks. If adopted, it could influence how other countries think about creator payment and AI training rules.

Should This Be A Requirement?

Experts share their thoughts…

Our Experts:

• Joe Z, Co-founder, DeAgentAI
• Othmane El Ouarzazi, mention.ma
• Karina Tymchenko, Founder, Brandualist
• Becca Greenberg, Founder and CEO, DeepThink Analytics
• Chris Sorensen, CEO, PhoneBurner and ArmorHQ
• Akhil Verghese, CEO, Krazimo

Joe Z, Co-founder, DeAgentAI

“I understand why people want AI companies to pay for training data, but honestly, I think that idea breaks more than it fixes. I’ve built products before, and once you turn learning itself into a toll road, only the biggest players survive. That’s not pro-creator, that’s pro-monopoly. If every sentence, image, or idea needs a license, you freeze open research and lock innovation behind lawyers.

“Models don’t store or resell content, they learn patterns, like humans do, just faster and messier. Charging per dataset sounds fair until you realize it pushes startups out and hands the future to whoever already owns the most data and cash. If you want creators supported, revenue sharing at the output layer makes more sense. I would say tax usage, not learning.”

Othmane El Ouarzazi, mention.ma

“If a business relies on search traffic, blocking AI crawlers is a mistake. Tools like ChatGPT, Google AI Overviews, and Perplexity are becoming where people start their search, so websites should treat them like search engines. I mean businesses should explicitly allow known crawlers such as GPTBot, Google bot, and Perplexity Bot to helps these systems understand the content and reference it correctly. When users look for products or services inside AI tools, that visibility can turn into real traffic and qualified demand.”

For any questions, comments or features, please contact us directly.

Karina Tymchenko, Founder, Brandualist

“I have learned from operating Brandualist that Content is not just raw Data; rather it is an outcome of years of experimentation with creativity, understanding your audiences and the subjective judgments of humans. The fact that when AI companies take years of that Content and use it to train models for the purpose of selling those results as output (AI) to customers without including the originators of the Content, destroys Trust within the Ecosystem entirely.

“I do not think a single blanket Fee is the Solution here. What is far more important is Consent and Fair Value Exchange. Brands and Content Creators need to be able to see how their Content is being used by AI Products and whether or not they are benefiting from its usage in any way. If there is no Balance between Innovation and the Content upon which the Innovation relies, at some point in time Innovation will overtake the Content upon which it was built.”

Becca Greenberg, Founder and CEO, DeepThink Analytics

“Whether companies must pay for content to train AI can’t easily be answered with a straightforward answer of “yes” or “no.” I believe the actual crux of the issue lies in questions regarding the transparency and level of consent that exist between those who own and control the distribution rights of training data sets and its end-users.

“The training data, in many cases, possesses significant value for AI developers, and thus, when large-scale use is made of it, it makes practical sense to create a framework for the development of licensing agreements and compensation plans for both parties involved. At the same time, if established fees are set too high, it may create barriers to entry and result in the entrenchment of the established, and most capable, AI developers.

“Therefore, going forward, it seems most likely that a hybrid approach will be needed whereby existing open data will continue to remain accessible to all developers, while additional tiers of data sets will be subject to more extensive contractual arrangements between the data source providers and the developers. As regulations become further entrenched in the industry, those who begin implementing governance tools today will be in a more favourable position.”

For any questions, comments or features, please contact us directly.

Chris Sorensen, CEO, PhoneBurner and ArmorHQ

“Well, I think first off charging AI companies for training on content is pretty reasonable but I think it needs to be handled carefully to avoid unintended consequences. Content creators deserve recognition and fair compensation when their work is used at scale, especially when it directly contributes to more commercial products. Saying that, I do believe overly rigid licensing or blanket fees could very much entrench the largest players who can actually afford them, which in turn hurts things for startups, open research, and even innovation.

“In my eyes, the more sustainable path is probably a tiered, transparent framework that distinguishes between public and domain content, as well as licensed datasets, and propriety or higher value creative work.

“I think the broad goal should not be to tax learning itself, but to better align incentives so that creators, platforms, and AI developers all benefit as the ecosystem fully matures over time.”

Akhil Verghese, CEO, Krazimo

AI models rely on the copyright work for training
“Enforcement of compensation for “copyright violations” during training relies on the government where the AI companies operate – even if American companies are forced to abide by this, Chinese companies with access to the same publicly available information will not be.

“Practically speaking, while it may be possible to compensate large content generators like the NYT or Reddit, it would be a herculean task to compensate every blog author whose work was used in the training of a state of the art model and deciding what portion is due.
Here are the major areas of debate

“Does the transformative nature under which the AI reuses the content constitute “Fair Use” in a legal sense.

“Does the difficulty in compensating everyone fairly just make the issue so impractical that it cannot feasibly be addressed.

“Specifically for America, where most of these companies are based, is dominance in AI over the next few decades so critical that the legal concerns above simply don’t matter.
Here are my personal opinions on each of these areas:
I’m not a lawyer, but based on my rudimentary understanding of the law and my slightly better understanding of how transformers work, I find it very difficult to classify the way AI models use training data as “Fair Use” in the traditional sense.

“I don’t think the difficulty in compensating everyone fairly justifies not making the effort. I don’t have an answer on how – it can’t be based on the amount of information, because that would encourage people to mass generate useless content on the off chance that it gets trawled. It can’t be based on outcome, because it’s basically impossible to attribute a particular response to a specific piece of information that may have adjusted a few model weights ever so slightly. But I do believe that if the brilliant minds that built these models dedicated some time to it, some sort of compensation structure might be possible.

“The need to stay at the forefront in AI is perhaps the most knotty problem of all. You want to feel like every company is operating under the same rules globally when it comes to creator content.

“Long story short – an ideal world solution would be for the top minds to get together and agree on some sort of compensation structure that is enforced at a global level. But I don’t think that is remotely likely to happen. Practically speaking, I feel the US sees the possibility of China overtaking them in the AI race as a very significant threat. While a few lawsuits may be settled in ways to compensate major information generators (like NYT and Reddit), I don’t think any enforcement will be enacted that significantly disadvantages US companies compared to Chinese counterparts.”