If your compliance diary says 2 August 2026 was the EU AI Act’s big deadline, it is now wrong in two directions at once. The heavyweight high-risk obligations you were bracing for have been pushed back more than a year. But the transparency duties — the ones that actually touch your website, your chatbot and your AI-drafted content — applied on schedule last month, and they are live now[1][2]. If you run digital for an association, society or university with members or users in the EU, the deadline that moved is probably not the one that affects you first.
What changed in July
The Digital Omnibus on AI — formally Regulation (EU) 2026/1744, adopted on 8 July 2026 — was published in the Official Journal on 24 July and entered into force on 27 July 2026[1][3][4]. It amends the AI Act’s timetable in two significant places. Obligations for stand-alone high-risk AI systems under Annex III — the category covering areas such as education, employment and biometrics — now apply from 2 December 2027, rather than 2 August 2026[2][3]. Obligations for high-risk AI embedded in regulated products under Annex I move to 2 August 2028[2][3].
That is welcome breathing room, and for most of our readers the honest summary is: the obligations you were least likely to trigger just became even less urgent. A membership platform or a conference website rarely operates a high-risk system in the Act’s sense. The systems that might qualify in academic settings — AI used in admissions decisions, exam proctoring, recruitment shortlisting — now have until December 2027, though the work behind them (risk management, documentation, governance) is slow to build and nobody sensible is treating the deferral as a pause[4].
What did not move
Article 50 — the transparency rules — applied on 2 August 2026, exactly as scheduled[5][6]. These are the duties most likely to reach a website you run, and there are four worth knowing.
First, chatbots must identify themselves. Providers of AI systems that interact directly with people must ensure users are informed they are dealing with AI, unless that is already “obvious to a reasonably well-informed, observant and circumspect person” in context[6][5]. Second, providers of generative systems must mark AI-generated or manipulated content in a machine-readable format so it is detectable as artificial[5][6]. Third, deployers — that is, organisations using these systems, not just the vendors building them — must clearly label deepfakes[5][6]. Fourth, AI-generated text published to inform the public on matters of public interest must be labelled, unless it has been through meaningful human review and editorial control[5][6].
Two details matter for planning. Systems already on the market before 2 August 2026 have a limited grace period, to 2 December 2026, for the machine-readable marking duty[5][7]. And the penalties are real: fines of up to €15 million or 3% of worldwide annual turnover, whichever is higher[5][6][7]. This is EU law, but it does not stop at EU borders — providers and deployers outside the EU are in scope where the system’s output is used in the EU, which is exactly the position of a UK, US or Australian organisation serving EU members[5][6].
What this means for a membership organisation
Translate those four duties onto a typical society or university website and the picture is manageable, provided someone actually looks.
If there is a chatbot on your site — member support, conference FAQs, a helpdesk widget — it needs to say it is AI. The formal duty sits mostly with the provider, but you chose the vendor and you configured the widget, so check what it actually displays. Ask your supplier where they stand on the Commission’s guidelines, published in July 2026 alongside a voluntary Code of Practice on transparency of AI-generated content[8][7]. A vendor who cannot answer that question is telling you something.
If your team uses AI to draft news items, newsletters or conference copy, the text-labelling duty has a carve-out you should take seriously rather than treat as a loophole: text that has been through human review and editorial control, with someone accountable for it, is treated differently from text published straight from the machine[5][6]. In our own AI Augmentation work the rule has been the same since before the Act required anything: a human stays on publish, and the system logs what was drafted by whom. That habit now doubles as compliance evidence.
Images and video deserve a closer look than text. If anyone in your organisation is generating imagery for campaigns or social media, the marking and deepfake rules apply to realistic synthetic content, and the tools you use should be embedding machine-readable marks[5][6]. Most reputable generators now do; confirm rather than assume.
Your January inventory is still the right list
In January we suggested a 2026 AI Act checklist for academic organisations: inventory your AI systems in one spreadsheet with a named owner, classify what you find, and deal with the duties already in force. Every line of that advice survives the Omnibus. What changed is the sequencing. The high-risk classification work now has until December 2027 for Annex III systems[2], but the inventory itself is what tells you whether any of the Article 50 duties apply to you today — and those are the ones with an enforcement date in the past.
The AI literacy duty also stayed. It has applied since 2 February 2025 on the Commission’s timeline[2], and the Omnibus softened its wording — organisations must now take measures to support a sufficient level of AI literacy among staff, rather than guarantee one[3][4]. For a small team, that looks like an hour of structured training and a short written note of who received it. Not onerous. Just not optional either.
So: cross out the deadline in your diary and write in two new ones. December 2027 for anything high-risk. And last month for the transparency duties — which is a polite way of saying the right time to check your chatbot is this week. An hour of your attention, one spreadsheet, and a couple of pointed questions to your vendors will cover most organisations. That is a better afternoon’s work than most regulations ask of you.
References
- European Parliament and Council. Regulation (EU) 2026/1744 amending Regulations (EU) 2024/1689, (EU) 2018/1139 and (EU) 2023/1230 (Digital Omnibus on AI). EUR-Lex, 8 July 2026. https://eur-lex.europa.eu/eli/reg/2026/1744/oj. Accessed 7 September 2026.
- European Commission. AI Act — Shaping Europe’s digital future. digital-strategy.ec.europa.eu, updated 3 August 2026. https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai. Accessed 7 September 2026.
- White & Case LLP. EU AI Omnibus enters into force, amending the AI Act. whitecase.com, July 2026. https://www.whitecase.com/insight-alert/eu-ai-omnibus-enters-force-amending-ai-act. Accessed 7 September 2026.
- Lewis Silkin LLP. The Digital Omnibus on AI enters into force today. lewissilkin.com, 27 July 2026. https://www.lewissilkin.com/insights/2026/07/27/the-digital-omnibus-on-ai-enters-into-force-today-102nedo. Accessed 7 September 2026.
- European Commission. Transparency obligations under Article 50 of the AI Act (FAQ). digital-strategy.ec.europa.eu, n.d. https://digital-strategy.ec.europa.eu/en/faqs/transparency-obligations-under-article-50-ai-act. Accessed 7 September 2026.
- Addleshaw Goddard LLP. AI transparency under the AI Act: what businesses need to know before 2 August 2026. addleshawgoddard.com, 2026. https://www.addleshawgoddard.com/en/insights/insights-briefings/2026/technology/ai-transparency-ai-act-what-businesses-need-know-before-2-august-2026/. Accessed 7 September 2026.
- Stibbe. The AI Act’s transparency obligations: rules, scope and timeline. stibbe.com, 27 July 2026. https://www.stibbe.com/publications-and-insights/the-ai-acts-transparency-obligations-rules-scope-and-timeline. Accessed 7 September 2026.
- European Commission. Guidelines on transparency obligations for providers and deployers of certain AI systems. digital-strategy.ec.europa.eu, July 2026. https://digital-strategy.ec.europa.eu/en/policies/guidelines-transparency-ai-generated-content. Accessed 7 September 2026.
