15 million euro buses: AI labeling for everyone – unless you read about it briefly.

Since August 2, 2026, the EUdSSR has mandated the labeling of AI-generated content. Anyone who has a text written by a machine and presents it to the public as information on politics, administration, health, or the environment must disclose that a machine was involved; otherwise, they face fines of up to €15 million or three percent of their global annual revenue. Unless, of course, someone briefly reads the fine. In that case, the obligation is completely waived—imposed and then revoked in the same paragraph.

Four cases, one label, one deadline
Article 50 of the AI ​​Regulation governs four scenarios. Chatbots must identify themselves as machines, generative systems must mark their outputs in a machine-readable format, emotion recognition must be disclosed, and deepfakes and AI-generated texts on matters of public interest require visible labeling. The regulation has been applicable since August 2, 2026.The maximum fine is 15 million euros or three percent of global turnover, whichever is higher.

Privately, you're exempt. But as soon as you regularly derive economic benefit from the activity, you're considered an operator and are fully liable. The baker with a newsletter, the sole proprietorship with a blog, the small publishing house with three employees: they're all included. The commission has interpreted what constitutes "public interest" broadly, encompassing everything from politics and administration to the judiciary, fundamental rights, health, and the environment, as well as any economic, scientific, or cultural development that could be the subject of public debate. In other words: everything.

The paragraph that withdraws itself
And then comes subparagraph two. The obligation This does not apply if the AI-generated content has been subjected to human review or editorial control and a natural or legal person bears editorial responsibility for the publication.No label, no note, no asterisk. The text can be 100% machine-generated, as long as a human with expertise has consciously reviewed and approved it.

In its response dated July 24, 2026, the Commission added that spell-checking is insufficient; a deliberate review of the content by a qualified person is required. This sounds strict. However, Article 50 contains neither a documentation requirement nor a rule regarding the burden of proof. There is no audit protocol, no register, no verification. There is simply an assertion that costs nothing.

Anyone who has an editorial team is out.
This clarifies who this regulation protects and who it affects. Publishing houses with editors-in-chief, corporate communications with approval chains, ministries with press offices, agencies with dual control: all exempt. Ironically, those very institutions whose machine-generated output streams daily onto millions of screens are allowed to remain silent because someone, somewhere, ticked a box. The individual, lacking an editorial structure, slaps on the label that the big players don't need. Anyone wanting to see how this redistribution of credibility works in other ways can find them here. in the transformation of critical journalism in its purest form.

To be fair, the opposing argument isn't foolish. Editorial responsibility means liability; a named individual stands behind the content with their name, and that's worth more than a sticky note that no one reads. A label identifies the source, responsibility identifies the culprit. However, this very argument undermines the regulation from within. If accountability is the true protected interest, then the label is merely decorative for everyone else. If the label is the protected interest, then exempting large corporations is a license to do whatever they want. You can't have both at the same time. Brussels decided on both simultaneously.

The voluntary code with emphasis
Then there is the Code of Conduct for the Transparency of AI-Generated ContentIt is explicitly voluntary. Those who sign enjoy legal certainty. Those who do not sign must demonstrate compliance using alternative, appropriate means and, according to the Commission, can expect more requests for information. Voluntary here means: voluntarily receiving more mail from the authorities.

Deadlines apply to you, not to Brussels.
The rest is administrative comedy. Digital Bus Regulation (EU) 2026/1744 It entered into force on July 27, 2026, and postponed the high-risk obligations from August 2026 to December 2027, product-embedded systems to August 2028, and national real-world laboratories to August 2027. The justification given was that the harmonized standards were not yet finalized. Member States, in turn, had until August 2, 2025, to implement their Market surveillance authorities are appointed, the German draft law on this was The Bundestag will not debate the issue until March 2026.If a citizen misses a deadline, it costs up to 15 million. If the state misses a deadline, there will be legal proceedings, eventually, perhaps. The same machinery that twenty million words rulebook Having been eliminated, she is unable to keep up with her own schedule.

And Switzerland? It watches, but not unscathed.
There is no AI legislation in this country. The Federal Council intends to ratify the Council of Europe's AI Convention and submit a draft consultation document by the end of 2026, sectorally, based on principles, without adopting the AI ​​Act. The relief lasts only until you realize that the regulation also applies to providers outside the Union as soon as their output is used within the EU. The Swiss blog with German readers is already fully affected, without ever having been consulted.

What has emerged here is not a transparency rule, but a system of jurisdiction: those who can demonstrate a hierarchy are credible, those who cannot are suspect. The regulation doesn't protect the reader from the machine; it protects the apparatus from the suspicion of using that same machine. And while Brussels punishes citizens for honesty with millions in fines, it postpones its own homework by sixteen months and calls it simplification. A label that is missing precisely where the reach lies is not a warning sign; it's a seal of approval for the apparatus! Whoever constructs transparency in this way doesn't want transparency; they want a hierarchy of who can speak unprompted and who has to identify themselves first!