Forum
Italy Writes Human Authorship Into Law as a Synthetic Actress Seeks an Agent

Two stories from the past fortnight approach the same question from opposite ends. What, exactly, does a human contribute to creative work made with AI?
On 17 September, Italy’s Senate gave final approval to a national law on artificial intelligence, signed on 23 September. Italy is the first member state of the European Union to adopt a comprehensive national framework designed to sit alongside the EU AI Act. Among its provisions is an amendment to Italian copyright law: works created with the help of AI are protected only when they result from the author’s own intellectual effort.
Ten days later, at the Zurich Summit, a film-industry event held alongside the Zurich Film Festival, producer Eline Van der Velden said that her studio was in talks with talent agents interested in representing Tilly Norwood, a fully AI-generated “actress” created by her company. She said an agency announcement could come in the coming months.
What the Italian law adds
In practice, Italy’s copyright clause restates a principle that courts and copyright offices elsewhere have already applied. The US Copyright Office, for example, said in January that prompts alone do not make someone the author of an output. What is new is that a national legislature has written the principle into statute in the AI era, in plain terms.
The law also addresses text and data mining. It confirms that copying for AI training, including for generative models, falls under Italy’s existing data mining exceptions, which allow rights holders to reserve their works. That is consistent with the European approach: training is permitted in principle, but rights holders can say no.
For creators working in Italy, or selling there, the message is clear. Protection depends on being able to show what the human did.
What the synthetic actress claims
Tilly Norwood sits at the other end of the spectrum. She is not a human performer using AI tools. She is a character presented as if she were talent, with the implication that a production could cast her in place of a person.
Debate about synthetic performers is not new, but the framing here is. An agent represents a person: their interests, their consent, their career. Applying that structure to a generated character blurs a line that matters well beyond film. If a character can be “represented”, who is the client, and whose work trained her face, voice and movement?
Why designers should care
The two stories map onto everyday design practice more closely than they seem.
- Credit. When a campaign uses a synthetic model or presenter, who is credited for the performance: the studio, the prompt writers, the animators, the people whose likeness and work trained the system?
- Contracts. Clients increasingly ask for avatars and generated spokespeople. Contracts need to say who owns the character, what it was trained on and what it may not be used for.
- Protection. Under laws like Italy’s, the parts of a project that carry protection are the parts a human authored. Studios that cannot document those parts may find their own work hard to defend.
Evidence of authorship
The practical lesson is documentation, again. If protection depends on human intellectual effort, then sketches, decision logs, edits and selection records stop being housekeeping and become evidence.
Competitions have been heading this way for some time. The International AI Award asks entrants to explain how AI entered their process and what choices they made, and its 2025 rules provided for a final round of live interviews in which finalists discuss their work with the judges. The 2025 winners are now published. A conversation about decisions is still one of the best tests of authorship we have.
The line worth holding
Italy’s law draws a line around human creativity. The Tilly Norwood pitch tests how far the market is willing to step over it. Designers do not need to choose between embracing AI and rejecting it. They do need to be precise about where their own contribution lies, and to resist framings that turn tools into talent and talent into training data.



