Ales Frelih

# The day the EU AI Act stops being optional

For most of its life so far, the EU AI Act has been a strange kind of law: on the books, widely discussed, and almost never enforced. Obligations existed, but the machinery to act on them was still warming up. That changes on 2 August 2026.

This is not another “new obligation” story. The rules you already know are mostly staying put. What arrives on 2 August is something quieter and more consequential: teeth.

## What actually switches on 2 August 2026

From that date, the European Commission and its AI Office gain their supervision and enforcement powers. In plain terms, they can:

– request your documentation,

– run technical evaluations of AI models,

– demand compliance and risk-mitigation measures,

– restrict or withdraw a model from the EU market, and

– issue fines — up to 15 million euro or 3% of global annual turnover for the relevant breaches.

Until now, a company could treat compliance as a paper exercise with little downside. After 2 August, the downside is real and named. The law did not get stricter last week. It got enforceable.

## The quieter deadline that lands today

There is also a date most businesses missed entirely: 27 July 2026, 18:00 CEST. That is the cutoff to be included in the first public list of signatories to the Code of Practice on Transparency of AI-generated content. Signing is the route to a presumption of conformity with your Article 50 transparency duties — which shifts the burden of proof toward the regulator instead of you. You can still sign after the cutoff; you just will not be in the first published cohort.

For a business with no legal department, that presumption of conformity is one of the cheapest pieces of legal cover on the table.

## What this means for a normal small company

Most SMEs are not providers of general-purpose AI models, so the heaviest GPAI enforcement is not aimed at you. But do not relax on that alone. The two obligations that do apply to ordinary deployers now sit inside an actively enforced regime:

– **Article 4 (AI literacy)** — law since 2 February 2025. If you deploy AI and have no record of training, that gap is now a gap someone can act on.

– **Article 50 (transparency)** — applies from 2 August 2026. Label your chatbots; mark your AI-generated content.

The shift is psychological as much as legal. “We will deal with it if someone asks” was always weak. From 2 August, it stops being a plan at all, because someone can ask.

## What I would do this week

Three moves, none of which need a consultant:

1. **Assemble your documentation.** If the AI Office — or an enterprise customer running due diligence — asked to see your AI measures, could you produce them today? If not, that is the week’s first job.

2. **Check what you ship for 2 August.** Any public-facing chatbot or AI-generated content needs a disclosure plan.

3. **Close the Article 4 gap.** A one-page policy and a dated training record. That is the difference between “exposed and hoping” and “documented and calm.”

The businesses that treated the AI Act as theoretical are about to find out it is not. The ones that quietly did the paperwork will spend August calm while everyone else scrambles.

**Not sure what applies to you? Start with the free 2-minute self-check → it tells you exactly which obligations are yours, then points you to the plain-language Handbook (€35) if you want the full walkthrough.** Link: linktr.ee/frelih

—

*This article was written by me and edited with AI — the kind of disclosure Article 50 is about. It is general information on the EU AI Act, not legal advice.*