Aug 2026enforcement live since
Feb 2025AI literacy live since
Aug 2025GPAI rules live since
2 Dec 2027high-risk duties (deferred)

In August 2026 a large part of the EU AI Act became applicable, and the supervision and penalty framework that sits behind it switched on. If your company builds, deploys, or simply uses AI inside the EU, the rules now have teeth. This is not a planning problem for later. It is the current state of the law.

The headlines could mislead you here. Much of the coverage focused on the high-risk system obligations that were originally due on that date. Those moved (more on that below). What did not move is the enforcement architecture, the penalties, and the obligations that already apply to ordinary businesses. Reading "the EU delayed the AI Act" and concluding you have more time is still the mistake we see most often.

The August 2026 milestone was less about a single new rule and more about the moment the whole Act became enforceable. Preparation stopped being optional.

Agentic Fluxus

The delay was narrow, not general. The Digital Omnibus, formally adopted on 8 July 2026, deferred the high-risk Annex III obligations to 2 December 2027. Article 4 AI literacy, the prohibited practices, and the GPAI rules were not moved.

What Actually Applies Since August 2026

The obligations that gained teeth

Enforcement, transparency, and literacy supervision

Three things matter for most EU SMEs now that this milestone has passed.

First, the enforcement and penalty framework is active. National market surveillance authorities now supervise and enforce, and that includes the Article 4 AI literacy duty that has applied since February 2025. Your obligations did not start in August 2026, but the ability to check and sanction them did.

Second, the transparency rules are in effect. If you run an AI chatbot, generate synthetic media, or use AI in ways a customer would want to know about, you need to disclose that clearly.

Third, the enforcement powers over general-purpose AI (GPAI) model providers are in force. The GPAI obligations themselves have applied since 2 August 2025, so if you develop or substantially fine-tune a model, you are already inside them. The Commission can now investigate and fine GPAI providers.

Does the live enforcement framework reach you?

Your staff use AI tools (Copilot, ChatGPT, embedded features) for work
You run a customer-facing AI chatbot or generate AI content
You develop or heavily fine-tune your own AI model
No one in the business touches any AI system, anywhere
You have no exposure to the EU market at all
Tip from Flux

Quick test: if AI touches your product, your staff, or your customers inside the EU, your compliance has to be demonstrable now, not just planned.

What Moved, and Why It Is a Trap

The high-risk deferral

Later, but not gone

The high-risk system obligations (Annex III: hiring, credit, education, essential services, and similar) were pushed from their original August 2026 date to 2 December 2027 under the Digital Omnibus, which was formally adopted on 8 July 2026. That makes 2 December 2027 a confirmed date to plan against, not a provisional one.

The trap is simple. The deferral bought time on one complex category. It did not pause the literacy duty, the prohibited-practice ban, the GPAI rules, or the enforcement framework. Companies that hear "delay" and stand down are choosing to be unprepared on the parts that did not move.

What To Do Now

A readiness baseline

Know your AI inventory

Every AI tool in use, including informal ones like ChatGPT

Close the literacy gap

Role-relevant AI training, documented, per Article 4

Fix transparency

Disclose chatbots and AI-generated content clearly

Keep records

Evidence you can show a market surveillance authority

Compliance Checklist

Readiness under live enforcement

Click to check off

Common Questions

A large part of the Act became applicable and the supervision and penalty framework switched on. National market surveillance authorities now enforce it, and the Commission gained enforcement powers over GPAI model providers.

Only in part. The high-risk Annex III obligations were deferred to 2 December 2027 by the Digital Omnibus, formally adopted on 8 July 2026. The AI literacy duty, prohibited practices, and GPAI rules were not moved and remain in force.

No. GPAI model obligations have applied since 2 August 2025. What started in August 2026 is the Commission's power to supervise and fine GPAI providers.

Yes. The AI literacy and transparency duties have no company-size threshold. A small deployer carries the same core obligations as a large one, scaled to context.

Free tool: unsure where you stand? Run the AI Readiness Check to see your exposure in about ten minutes.

What To Do Right Now

10 min
Run the free AI Readiness Check to see your exposure.
1 hour
Inventory the AI tools your staff already use.
3.5 hrs
Train your team to Article 4 with the Agentic Fluxus staff course.

Ready to get your team compliant?

Staff AI Compliance Awareness · €39/person · Certificate included · 14-day guarantee

View Course