What You actually Own when you use paid AI

Paying for an artificial intelligence service is not the same as purchasing the rights to the outputs that service produces. It is a distinction that many professional users have not yet internalised, and one that can have concrete consequences the moment those outputs are used in a commercial context.

The market for generative AI platforms has grown rapidly, and with it the variety of licensing models adopted. Understanding what you are actually buying — and what you are not — has become an integral part of the work for anyone using these tools on commission.

The parallel with Stock photography

For those working in communications, the system of image rights is familiar territory. The two fundamental categories — Rights Managed and Royalty Free — have defined the conditions of use for purchased content for decades.

A Rights Managed image is licensed for a specific use: a channel, a territory, a time period. Outside those limits, a new licence is required. A Royalty Free image, by contrast, can be used multiple times without paying royalties for each use, but remains subject to the conditions established by the licence — which is never an absence of rights, but a broader form of concession.

The fundamental difference between the two models is not about cost: it is about the rights transferred. In both cases, the image belongs to the author or the agency distributing it. The buyer obtains a licence to use, not ownership.

With generative AI platforms, the structure is analogous — but less transparent.

What the Terms of Service actually say

Every AI platform handles rights over outputs differently, and terms of service change over time. It is worth examining some concrete cases.

OpenAI states that users retain the rights to outputs generated through its APIs, provided they comply with the usage policies. However, the company also reserves — unless users opt out in paid versions — the right to use content uploaded as input to improve its models. Rights over the output, then, but with conditions on the input.

Midjourney has adopted different positions over time depending on the subscription plan. In free versions, generated content is subject to a non-commercial Creative Commons licence. In paid plans, commercial use is permitted, but the platform retains a broad licence to use the outputs for its own purposes. The distinction between “I can use it” and “I own it” remains significant.

Adobe Firefly has chosen an explicit positioning on the matter: the system is trained exclusively on commercially licensed or public domain content, and Adobe offers an indemnity guarantee for commercial use of the outputs. It is a different model, built around legal certainty as a product value.

Three platforms, three approaches. None of which corresponds exactly to the direct purchase of content with clear, fully transferred rights.

The question of Authorship

There is a further complexity, concerning not the contractual terms but the regulatory framework on copyright.

In many jurisdictions, including European and US law, copyright requires an identifiable human author. The United States Copyright Office has explicitly excluded from protection content generated by AI without a significant creative human contribution. The position of European authorities is evolving, but the general trend is analogous.

This means that an image generated entirely by a prompt, without documentable creative intervention by the user, may not be protectable as a copyrighted work. Anyone could reproduce or modify it without incurring infringement.

For a company using AI content in advertising campaigns, institutional materials or commercial products, this absence of protection is relevant: what is produced may not be exclusive in the legal sense of the term.

What you are actually Buying

To simplify, paying for a subscription or API access to a generative AI platform generally purchases three things:

• The Ability to Generate Outputs

  access to the model and its computational power.

• A Licence to Use the Outputs

  more or less broad depending on the plan, but almost never equivalent to full ownership.

• The Conditions of Use for Inputs

  which define what the platform can do with the data uploaded by the user.

What is not purchased, in most cases, is absolute legal certainty over the commercial use of outputs, a guarantee that generated content does not infringe third-party rights, and protection equivalent to that offered by content created and licensed through traditional methods.

The implications for those Working on Commission

For a communications agency or a creative professional using AI tools in the production of materials for clients, these considerations translate into concrete responsibilities.

The client commissioning a campaign, a catalogue or digital content expects to receive materials they can use without risk. If those materials include AI-generated outputs, the chain of rights must be verified with the same care reserved for any other content — photographs, illustrations, music.

The fact that a tool is accessible, powerful and widely adopted does not automatically make it safe from a rights perspective. As in the stock photography world, the difference between correct use and legal exposure depends on a precise understanding of what the licence permits — not on the assumption that “if it is available, I can use it”.

The complexity of the current landscape is not set to simplify any time soon. Platforms update their terms, regulations evolve, and courts are still defining the contours of copyright as applied to AI. In this context, relying on professionals who know the territory — and who take responsibility for the choices made — is a concrete form of protection for any company wishing to use these tools without unexpected exposure.