Skip to content
PenzaıHUMAN TECH

The AnalysisEurope

Labeling AI-generated content: what a Genoese notary of 1154 can teach you

Since 2 August 2026 the AI Act requires AI-generated content to be flagged. What it means for a small business, what it does not, and how to keep a record.

The Professor at PenzaiReviewed by Agostino Cacciola6 min read

Un team di marketing in ufficio guarda schermi con immagini e testi di una campagna in preparazione

You have published an AI-made image on your website or social media, and nobody ever asked how you produced it. In Europe that question may now come: Article 50 of the AI Act is all about synthetic content. The Professor wants to work out with you who has to do what, because for a small business the answer is less dramatic than it looks, but it is not "nothing" either.

What Article 50 says, no embroidery

The text is in Regulation (EU) 2024/1689, also available on the Commission's AI Act Service Desk. The transparency rules apply from 2 August 2026. Two paragraphs matter to us.

  • Paragraph 2, for whoever provides the system. Providers of AI systems that generate synthetic audio, images, video or text must ensure the outputs are marked "in a machine-readable format" and detectable as artificially generated or manipulated. The technical solution must be effective, interoperable, robust and reliable as far as technically feasible. The duty does not apply where the system only performs an assistive function for standard editing and does not substantially alter the input data or its meaning.
  • Paragraph 4, for whoever uses the system. A deployer who publishes a deepfake, meaning an image, audio or video generated or manipulated by AI, must disclose it. For evidently artistic, creative, satirical or fictional works, a notice that does not hamper enjoyment of the work is enough. It also covers AI-generated text published to inform the public on matters of public interest, unless there is human review or editorial control and a person or entity holds editorial responsibility.

A note on dates. Regulation (EU) 2026/1744, the postponement we covered in AI Act delay to 2027, added a transitional rule to Article 111: providers of systems already on the market before 2 August 2026 must take the necessary steps to comply with paragraph 2 by 2 December 2026. Time is running out, but for providers, not for you.

The medieval case: Giovanni Scriba's register

Let's jump to Genoa. It is the mid-twelfth century and the city has understood that trade does not live on ships alone: it lives on paper, or rather on trust put in writing. The Museums of Genoa hold the cartulary of Giovanni Scriba, described as the oldest surviving notarial register and the oldest paper codex in Western Europe: 196 folios, kept in the State Archive. Scholars date it from 1154; Calleri's research on Ligurian notaries takes it as the oldest to survive.

What was inside? The deeds of a merchant city opening to the Mediterranean. Giovanni, who was also a scribe of the Commune of Genoa, recorded contracts as they were made. The point, and the Professor allows himself a small digression, is that the register was not there to look good. It was there so that, when two merchants quarreled, someone could say "here is what was said, before whom, on which day".

And you will ask: what does a notary of 1154 have to do with an AI-made image? The problem is identical. When anyone can produce a credible document, photo or text, the value is no longer in producing it: it is in the trail that shows where it came from. The notary did not guarantee the contract was fair; he guaranteed it had existed, in that form, at that moment.

Well, your company today sits in that same workshop. The register is yours to keep.

What actually falls on a small business

Most of the time a small business is a deployer: it uses other people's tools. You do not have to build the marking system; that is the job of whoever provides the generator. It is up to you, though, not to strip or get around the signals, and to respect paragraph 4 when you publish.

SituationWhat to do
Product photo retouched with AI (background, lighting)Usually assistance with editing; note the tool used
Realistic but invented person in a campaignDisclose it clearly: the typical case of content that looks authentic
Video with a cloned voice or face of a testimonialNeeds the person's consent and disclosure; without them, don't
Evidently creative or fantasy illustrationA discreet notice is enough
Blog post written with AI and reviewed by a personWith human review and someone taking responsibility, paragraph 4 on text does not apply
Public-interest press release generated by AI without checksDisclose it; better still, have a person review it

The table is a guide, not legal advice: for borderline cases, read the text and talk to whoever advises you on the law. Other rules apply already: misleading advertising, image rights and the GDPR. On copyright and crawlers, see robots.txt and AI training.

Content provenance: what is emerging

Paragraph 2 does not say which technology to use. Worldwide, though, an open standard exists for describing a file's origin: it is developed by C2PA, the Coalition for Content Provenance and Authenticity, a project of the Joint Development Foundation working on open technical specifications for content provenance and authenticity. In practice: signed metadata that travels with an image and tells whoever reads it how it was produced.

The Professor's advice, from an old librarian: don't make it a religion. These signals can be lost when a file is recompressed or re-uploaded to a social network, and not every tool writes them. Your internal register remains your first line of defense.

Your own "cartulary" in the company

You don't need software. A shared sheet with these columns will do:

  • What: the content (image, text, video) and where it was published.
  • How: the tool used, the date, who produced it.
  • How much AI: generated from scratch, retouched, only corrected.
  • Who checked: the name of whoever gave final approval.
  • Label: whether and how it was flagged to the public.

Typical scenario, not a real client: a travel agency of 8 people uses an image generator for its package holidays. A customer writes that the beach in the picture doesn't exist. With the register, in two minutes the agency knows when it was created and by whom, and can fix it or add a notice. Without it, it loses half a day reconstructing and some trust. The same logic applies when content ends up in your sales channels, which we discuss in Artificial intelligence for marketing.

What to do on Monday morning

  1. List where you publish AI-made content. Website, social media, newsletter, flyers, presentations. One line per channel and tool.
  2. Create the register sheet with the columns above and have the next ten pieces of content filled in. Decide who gives final approval.
  3. Write a simple rule about people. No cloned real faces, voices or names without written consent, and a clear label when a realistic image is synthetic. If you have no internal rules yet, start from ChatGPT at work: rules for employees.

Giovanni Scriba did not write for posterity: he wrote because the next day someone would ask him to account for it. The same goes for us. If you want to put your company's content in order without bureaucracy, book a free introductory call: in half an hour we look at what you publish, what you must flag and how to keep track. You can also see our services and the marketing page.

The Professor of Penzai

Was this useful?
Buy us a coffee

Penzai Magazine

The day's stories, every evening in your inbox

Briefings, guides and analyses from the Penzai Professor: a preview and link for each article, one email a day. Free gift: the AI checklist for SMEs. Unsubscribe anytime.

Prefer Telegram? Get every article as soon as it is out Follow on Telegram

Comments

No comments yet. Be the first.

All articles
Portrait of the smiling Penzai mascot

Let's talk about your next project.

A focused consultation to understand how technology can really help you. First we understand your work, then we find the right technology.