Graphic design in the age of AI

These are the questions the field is still working through. They are framed here as open, because they are.

Training-data copyright. On 13 January 2023, artists Sarah Andersen, Kelly McKernan and Karla Ortiz filed a class action, Andersen v. Stability AI, against Stability AI, Midjourney and DeviantArt. It centred on the LAION image set used to train the models. In February 2023, Getty Images sued Stability AI in the US (with a parallel UK action) over the scraping of millions of its licensed images. Both tested, rather than settled, how copyright applies to model training. The Andersen case was still in active litigation, with no final ruling on the merits, well into 2026. The UK strand reached judgment first: on 4 November 2025 the High Court rejected Getty’s copyright claims, holding that a model’s trained weights are not “infringing copies” — an early, jurisdiction-specific result rather than a general answer.

Authorship and registration. On 21 February 2023, the US Copyright Office ruled on Zarya of the Dawn, a graphic novel by Kris Kashtanova: the Midjourney-generated images were held not protectable, while Kashtanova’s text and the selection and arrangement of elements were. On 16 March 2023, the Office issued formal registration guidance — copyright protects only human-authored material, purely AI-generated output is not registrable, and more-than-de-minimis AI content must be disclosed on an application. In March 2025, the US Court of Appeals for the DC Circuit affirmed that principle in Thaler v. Perlmutter: a work authored entirely by a machine, with no human author, cannot be registered — the first appellate decision to settle the point in US law.

A licensed-data response. In March 2023, Adobe announced Firefly, built into Photoshop and Express. Its first model was trained on Adobe Stock, openly-licensed content and public-domain material, and was positioned as “commercially safe,” a direct answer to the training-data questions hanging over the open models. The positioning was not unchallenged: reporting in April 2024 said a small share of Firefly’s training images were themselves AI-generated (including Midjourney output), which complicates a clean “licensed-data only” claim. Whether “trained on licensed data” fully resolves the authorship and ownership questions is itself still debated.

It is too early to write the legacy of a shift this is still happening inside. What can be said honestly is narrower.

The tools are now part of the standard toolchain, and a working designer will encounter them whether or not they seek them out. The legal and authorship questions are live, not closed — a studio that builds a deliverable on purely AI-generated output is building on unsettled ground, and the prudent response is to know that rather than assume it away.

The skills that survive the shift are the ones the prompt does not replace: judgement about what a piece is for, the ability to read a brief and a client, typographic and compositional discrimination, and the editorial sense to know when an output is right and when it is merely plausible. A design education teaches exactly those things, and they become more load-bearing, not less, when the production step gets cheap.

Whether AI ultimately deskills the field or raises the floor under it is a genuine open question, argued in both directions by serious people. This page does not take a side on it. It records what happened, when, and from sources you can check.