General information, not legal advice.
Breaches of the EU AI Act's transparency rules for AI-generated content (Article 50), including failing to mark AI images or to visibly label deepfake-class content, can be fined up to 15 million euros or 3 percent of worldwide annual turnover, whichever amount is higher. The rules have applied since 2 August 2026, enforcement is handled by national authorities, and both the platforms that generate images and the brands that publish them can be fined for breaching their respective duties. That is the direct answer; the rest of this page is the context that makes it useful.
Where Article 50 sits in the penalty structure
The AI Act grades its fines in tiers, and knowing the ladder stops both panic and complacency:
| Breach | Ceiling | Or % of turnover |
|---|---|---|
| Prohibited AI practices (the banned list) | 35 million euros | 7%, whichever is higher |
| Most other obligations, including Article 50 transparency | 15 million euros | 3%, whichever is higher |
| Supplying incorrect or misleading information to authorities | 7.5 million euros | 1%, whichever is higher |
Unmarked AI content is a middle-tier breach: serious enough to carry a nine-figure ceiling for a large group, and clearly distinguished from the top tier reserved for prohibited practices. Note the third row on the way past: answering an authority's information request badly is its own fineable offence, which is one more reason the answer should exist before the question does.
Who can be fined
Both roles in the transparency system carry their own exposure. Providers, the platforms and tools that generate AI content, for failing the machine-readable marking duty. Deployers, the brands, retailers and agencies that publish it, for failing the visible-labelling duty on deepfake-class content. One does not shield the other: a perfectly marked file published without its required label is the publisher's breach, and an unmarked file is the platform's, whoever ends up embarrassed.
The SME rule most coverage misses
For small and medium-sized enterprises, including startups, the Act flips the formula: the fine is capped at the same percentages or amounts, but whichever is lower. For a small fashion brand, 3 percent of turnover is a much smaller number than 15 million euros, and that is the ceiling that applies. The exposure is real but proportionate to size, which is worth knowing before a vendor uses the headline figure to sell you urgency, ourselves included.
What actually determines the amount
Ceilings are not price lists. Authorities set fines within them weighing, among other factors: the nature, gravity and duration of the breach; whether it was intentional or negligent; remedial action taken; cooperation with the authority; and any previous breaches. In practice the pattern of every comparable European regime holds here: cooperative companies with documentation and prompt remediation meet orders and modest penalties; obstruction and recidivism meet the ceiling conversations. The variables above are also the only ones fully under your control, which is the actual takeaway of this page.
Quick answers
- Is the first contact a fine? No. The path runs complaint, information request, proceeding; fines sit at the end of it, not the beginning. The walkthrough is here.
- Can a brand be fined for images its platform generated? Yes, breaching their labelling duty as deployers, independently of anything the image platform did or failed to do.
- Do old images count? No. Content generated before 2 August 2026 never needs retroactive marking.
- Since when is any of this enforceable? Since 2 August 2026, with a narrow transition window for some existing AI systems until 2 December 2026, which never covered publishers' labelling duties.
Whether your organisation could answer the four questions a letter would ask is a three-minute check, the vendor half of the answer is ten questions long, and the full legal picture lives in the Article 50 guide.
Every image exported from Modelia carries signed provenance and an invisible watermark by default, and anyone can check any image, free, no account, in our public C2PA checker. The full picture, rules and industry data, is in The State of AI-Generated Content in Fashion 2026.
