For two years, 2 August 2026 was circled in every compliance calendar in Europe as the day the AI Act's heavy machinery switched on. Then, weeks before the deadline, Brussels blinked. The so-called Digital Omnibus agreement pushed the hardest obligations back, but it did not push everything back, and the parts that did arrive this month are precisely the ones that touch brands using AI for marketing, imagery and customer service.
What moved, and what did not
Under the Omnibus agreement, the obligations for high-risk AI systems have been deferred: stand-alone high-risk systems under Annex III now come into force on 2 December 2027, and AI embedded in regulated products under Annex I follows on 2 August 2028. For most home and garden suppliers this matters less than the headlines suggest, because a product imagery pipeline or a customer service chatbot was never a high-risk system in the first place.
What did arrive with 2 August 2026 is the part of the Act aimed at everyday commercial AI use: the transparency obligations under Article 50, the enforcement powers over general-purpose AI models, and the full penalty regime behind them. In other words, the rules that were designed for chatbots, generative AI content and synthetic media are now live, with teeth.
One caveat worth knowing: at the time of writing the Omnibus still needs its formalities completed. If its adoption slipped past 2 August, the original timetable would technically apply as written until it lands. That is a lawyer's detail, but it is a good reason not to treat the deferrals as an excuse to do nothing.
The three rules that touch a retail brand
1. Tell people when they are talking to a machine
If your webshop runs an AI chat assistant, customers in the EU must be able to tell they are dealing with AI. In practice this is a disclosure line and sensible design, not a rebuild. If your chat tool is bought in, your vendor should already handle it, but the brand carries the customer relationship, so check rather than assume.
2. Synthetic content should be identifiable as synthetic
Article 50 requires providers of generative systems to ensure AI-generated content carries machine-readable marking, and it requires clear disclosure for deepfakes, meaning AI content that could pass as authentic footage of real people, places or events. A lifestyle render of your pergola in a summer garden is not a deepfake of anyone, but it is synthetic content, and the direction of travel is unmistakable: provenance is becoming part of the supply chain, the same way pack data and safety certificates already are.
3. The penalty regime is live
The fines scale to global turnover. Nobody is fining a garden brand for a lifestyle render this autumn, but "we did not know" stops being a defence once the enforcement powers exist. What regulators and, more practically, retail customers will increasingly ask for is simple: can you show where your content came from?
Does this reach UK brands?
The Act applies to AI systems and outputs placed on the EU market, wherever the company sits. If you sell into the EU, run EU-facing webshops, or syndicate product content to EU retailers and marketplaces, the transparency expectations follow your content there. If you sell only in the UK, the Act does not bind you, but the UK's own advertising and consumer rules very much do, and we have written a separate plain-English guide to where UK law actually stands on AI imagery in 2026.
Provenance is becoming part of the supply chain, the same way pack data and safety certificates already are.
What we would do this quarter, if we were you
- Inventory the touchpoints. List where AI already meets your customers: chat widgets, generated product imagery, AI-written copy, video. Most brands are surprised by the length of the list.
- Sort your disclosure posture. A short, confident line beats a buried footnote. We are open about our own methods for a reason: customers reward honesty and regulators require it.
- Ask your production partners about provenance. Any serious AI imagery supplier should be able to tell you exactly which images were generated, from which references, when, and with what marking. If they cannot, that is your answer.
- Keep the reference chain. The strongest protection in product marketing was never an AI rule at all: the image must truthfully show the product being sold. Build every asset from the real product and you satisfy consumer law and the spirit of the AI Act in one move.
- Diarise the real deadlines. December 2027 and August 2028 are when the high-risk machinery lands. If your business also builds or embeds AI systems, that planning starts now, not then.
Every image we produce is built from the client's real product reference, logged against its source files, and delivered with a full production record. We are openly an AI studio, it says so on our homepage, and that transparency is exactly the posture the new rules reward.
This article is general information for busy brand teams, not legal advice. For decisions about your own compliance, speak to a solicitor who knows AI and advertising law. Sources: Gibson Dunn on the Omnibus agreement, Travers Smith on the delayed deadlines, aiacto on what changes on 2 August 2026, DLA Piper on the deferral, the AI Act high-level summary.