What the Nemo Plus AI dispute reveals about AI-generated content, AI tools, ownership and liability.

In late 2024, users of Figma, the cloud-based collaborative design platform beloved of creative studios, brand agencies and UX teams worldwide, discovered what some described as unsettling in their administrative settings. According to a proposed class action filed in a California federal court, a toggle feature labelled “Content Training” had, for certain subscription tiers, allegedly been switched on by default. The complaint alleges that customer designs, prototypes and other creative content could therefore be used to train Figma’s Artificial Intelligence (AI) products. The plaintiffs claim that owners of the works used for that purpose had not provided informed consent.

The Figma class action is more than a cautionary tale from America. It highlights a growing challenge for businesses operating in the creative industry as AI tools become increasingly embedded in design, marketing and content creation workflows.

The lesson is not that creative businesses should avoid AI tools. It is that every AI tool has boundaries – and those boundaries are usually found in the small print. If you don’t understand what your AI tool allows, restricts, relies on, and guarantees, you can’t confidently define the promises you make to your own clients.

RELATED: Feeling the Music: The case that put AI music recommendations to the patent test

An Ancient Maxim for Frontier Technology

People have been overpromising for as long as they have been making promises – and mothers, lawyers and other sound counsel have been warning them against it for just as long.

Between 530 and 533 AD, Emperor Justinian established a commission to collect and compile centuries of juristic writing, producing what became known as Justinian’s Digest. Within this compilation, the fifty-fourth passage of the seventeenth title, De diversis regulis iuris antiqui, we find the memorable maxim: nemo plus iuris ad alium transferre potest, quam ipse haberet — no one can transfer to another more rights than they themselves possess.

The principle resonates with particular force in the age of generative AI. A creative business that promises its client original, non-infringing work and assigns all intellectual property in its deliverables cannot give more than it has. Where the AI tool’s terms do not clearly vest the relevant rights in the user, or where copyright does not subsist in the AI-generated output at all, such client-facing assurances may require careful qualification.

Absent such qualifications, creative businesses risk finding themselves in breach of warranties, exposed to IP claims, or carrying liabilities that the AI tool provider has carefully avoided.

The Creative Promise and the Provider Promise

Let us turn to the promises.

A creative studio enters into a services agreement with its client. That agreement typically requires it to:

  • assign or vest all IP in deliverables to the client;
  • warrant that deliverables are original and do not infringe third-party IP;
  • indemnify the client against IP infringement claims;
  • maintain confidentiality over client information and brief materials; and
  • accept liability – sometimes uncapped for IP breaches – if things go wrong.

The same studio is likely to subscribe to one or more AI tools: a generative design platform, a large language model, an image generator, a stock media library with AI features, or an autonomous agent. The tool provider’s terms may:

  • assign output IP only “if any” and “to the extent permitted by applicable law”;
  • expressly disclaim warranties of non-infringement;
  • provide qualified non-infringement indemnities, but exclude or cap liability at modest amounts;
  • reserve rights to use inputs and outputs for model training or service improvement, depending on tier and settings; and
  • impose commercial-use or acceptable-use restrictions that may limit how outputs are deployed.

AI Tools: The Gap

The risk lies in the gap between these two sets of promises. The studio has made an uncapped or unqualified commitment to its client. By contrast, the tool provider has given the studio a qualified, capped and disclaimed set of permissions.

Not all AI tools create equal risk. They may be used for ideation and research, generative design and image creation, coding and prototyping, stock-media enhancement or autonomous agentic tasks. The risk exposure is contextual and depends on how the tool’s output feeds into the client deliverable.

There is, however, a discernible pattern: the cheaper and more consumer-facing the tool, the more likely it is that the user carries the risk. Paid enterprise and API tiers often improve the position materially, but they do not eliminate the need to read the terms carefully. Their terms may still contain qualified indemnities, liability caps, residual data-use rights and acceptable-use restrictions.

A further risk factor for South African users is jurisdictional. Many AI tools are governed by United States – often Californian – or European law. Local creative businesses may therefore be using tools designed, priced and risk-allocated for foreign markets, while simultaneously giving clients commitments governed by South African law.

Read the Small Print, then Map the Gap

All of this makes it essential to understand precisely what one’s AI tools permit, for careful alignment between those permissions and the commitments made to clients. This is achieved through contract diligence and a structured gap analysis.

Key diligence areas include:

  • IP ownership and assignment;
  • non-infringement warranties and indemnities;
  • limitations of liability;
  • confidentiality and data-use rights;
  • commercial-use restrictions; and
  • acceptable-use policy restrictions.

Read the small print. Do the gap analysis. Align your promises with your permissions.