Digital Omnibus AI Act: high-risk delayed – what SMEs must still finish by 2 August 2026
On X and LinkedIn, one line keeps circulating: “The AI Act was postponed.” That is only half true. The Digital Omnibus on AI delays key high-risk obligations – but the transparency rules most SMEs actually touch still land on 2 August 2026.
If you run chatbots, AI agents, lead workflows or content pipelines, do not treat the relief as a free pass. This guide separates headline from reality – and turns both into a short implementation checklist for automation teams.
Executive summary
- High-risk: stand-alone Annex III systems expected from 2 December 2027; product-embedded systems from 2 August 2028.
- Article 50 transparency: still largely applies from 2 August 2026 – chatbots, deepfakes, certain AI content.
- Article 4 AI literacy: already in force since February 2025; the delay does not change that.
- For automations: build inventory, disclosures, approvals and audit logs now – not in 2027.
- LinkedIn-ready takeaway: time gained on high-risk, deadline remains on transparency.
What the Digital Omnibus actually delays
At the end of June 2026 the Council gave final approval to the Digital Omnibus on AI. The aim: cut red tape, wait for standards, and ease the burden on SMEs and small mid-caps. The loudest change is the high-risk timeline shift:
| Area | Original | After Omnibus |
|---|---|---|
| High-risk Annex III (stand-alone) | 2 August 2026 | 2 December 2027 |
| High-risk in regulated products | 2 August 2026 | 2 August 2028 |
| Article 50 transparency | 2 August 2026 | stays (with narrow transition details) |
| Article 4 AI literacy | since Feb 2025 | unchanged |
Simplifications once reserved for SMEs are also being extended to small mid-caps – lighter documentation and sandbox access. Helpful for growing firms, irrelevant as an excuse to ignore August transparency.
What still applies on 2 August 2026
For most mid-market companies, Article 50 is the part that bites now. As soon as people interact with an AI system – or you publish AI-generated or heavily manipulated content – you need clear disclosure. Practical examples live in our guide to the EU AI Act labelling rules 2026.
AI interaction
Website chatbot, WhatsApp assistant, phone AI: users must recognise they are talking to AI – unless that is already obvious.
Published content
Deepfakes and certain AI texts on matters of public interest need visible labelling; provider-side machine-readable marking stays on the radar.
The most common mistake right now
“High-risk is 2027 → we can wait.” That is how chatbot, content and shadow-AI gaps appear in August. The delay buys time for heavy compliance stacks – not for basic transparency.
What this means for AI automations and agents
Lyron builds one thing for mid-market teams: processes that run reliably – with n8n, Make, Power Automate or custom software. Compliance belongs in the same design as triggers, permissions and error handling.
- 1. Inventory before features: Which system talks to customers? Which output goes public?
- 2. Disclosure as a default step: chat entry, call start and content publish get fixed transparency blocks.
- 3. Human-in-the-loop where risk rises: approval before send for sensitive cases – as a workflow node, not an afterthought.
- 4. Evidence by default: log model version, approval, timestamp and final output – otherwise the audit trail is missing.
If shadow tools are still spreading, read our guide on shadow AI in companies. And if teams still need training: the Article 4 AI literacy obligation is not an August surprise – it already applies.
14-day checklist to August
Days 1–3
Inventory and priority
- List every customer-facing AI system (chat, phone, forms, content)
- Record the role per system: deployer, provider, or both
- Flag high-risk candidates – without tying the August deadline to that flag
Days 4–8
Put transparency into the workflow
- Define standard wording for chat, voice and publications
- Place notices where the interaction starts
- Add approval and label steps to publish/send workflows
Days 9–14
Test, train, evidence
- Run test cases for service, marketing and sales
- Document a short training for affected teams
- Store an audit record: system, owner, label, approval, timestamp
FAQ
Did the Digital Omnibus postpone the entire EU AI Act?
No. The main delay covers high-risk duties: to 2 December 2027 for stand-alone Annex III systems and to 2 August 2028 for systems embedded in regulated products. Article 50 transparency obligations still largely apply from 2 August 2026.
What must SMEs actually implement by 2 August 2026?
For most mid-market companies that means transparency: disclose AI interactions, label deepfakes and certain AI-generated content, and document approvals. The Article 4 AI literacy duty has already applied since February 2025.
Does the delay also cover customer-service chatbots and AI agents?
A typical service chatbot rarely becomes high-risk solely because it answers customers. The duty to inform people they are interacting with AI remains relevant on 2 August 2026, regardless of the high-risk postponement.
What changes for automation projects built with n8n or Power Automate?
The platform itself does not change. Organisationally you need an AI inventory, clear roles, visible disclosures where users are involved, and audit logs for approvals. Those controls can be wired directly into n8n, Make or Power Automate workflows.
Is the Omnibus already binding law?
The Council gave final approval on 29 June 2026 and the act was signed in early July 2026. Until publication in the Official Journal, the original timetable remains the safe planning baseline. Prepare for Article 50 on 2 August 2026 and watch the final texts in parallel.
Sources and status
Article status: 17 July 2026. Key references:
- Council of the EU: final green light for the Digital Omnibus on AI (29 June 2026)
- Regulation (EU) 2024/1689 – EU AI Act
- European Parliament: approval of the simplification measures
Note: This article is practical guidance and not legal advice. Borderline cases should be reviewed by qualified counsel.
August-ready in two weeks?
We review your chatbots, agents and publish workflows – and wire disclosure, approval and audit logging directly into the automation.
Book a free consultation