If you've been bracing for a wall of AI compliance in August 2026, you can partly exhale. In a late-2025 Omnibus package, the European Union did something unusual: it postponed the hardest parts of its own AI Act. But postponed isn't cancelled, and two obligations still arrive in 2026 that quietly affect ordinary businesses — not just AI labs.
What got pushed back
The headline change: the rules for high-risk AI systems — the heavily documented category covering hiring, credit, biometrics and critical infrastructure — moved from 2 August 2026 to 2 December 2027. AI baked into regulated products (lifts, toys and the like) slips to August 2028. The official reason is readiness: standards and guidance weren't finished, and a compliance deadline without a rulebook helps no one.
For most SMEs, that's breathing room on the scariest, most paperwork-heavy tier. But don't mistake the delay for a reprieve on everything.
What still lands in 2026
Two things survive the delay, and they're the ones most likely to touch you:
- Chatbot transparency (August 2026). If a person is interacting with an AI system, they must be told. A chatbot, a voice agent, an AI receptionist — it has to be clear the user is talking to a machine, not a human. No more pretending.
- AI-content labelling (December 2026). Synthetic media — AI-generated images, audio, video and meaningful text — must be marked as such, in a machine-readable way (think watermarking and metadata). The deferral here was trimmed to just a few months.
Add to that the obligations already in force since 2025 — the ban on unacceptable-risk practices and a baseline AI-literacy duty for staff who use these tools — and the shape of 2026 is clear: not heavy certification, but honesty and disclosure.
What this means in practice
None of this requires a legal department. For a typical small or mid-sized business, the 2026 checklist is short and doable:
- Label your bots. Make sure every chatbot and voice agent clearly states it's an AI assistant. (If you built yours well, this is a one-line change.)
- Mark your AI content. Where you publish AI-generated images or media, keep the provenance metadata intact and disclose where appropriate.
- Train your team. A short internal briefing on responsible AI use covers the literacy expectation and prevents the dumb mistakes.
- Map what you use. Keep a simple register of the AI tools touching customers or decisions. If any drift into high-risk territory, you now have until December 2027 — use it.
The bigger lesson of the Omnibus is strategic, not legal. Europe extended its deadlines but held its line: AI you deploy must be transparent, and people must know when they're dealing with a machine. The businesses that treat that as a trust feature — not a burden — will find their customers prefer it.
