Video summary

15 MILLIONEN Strafe! Neue KI-Pflicht ab 2. August – das trifft JEDEN.

Main summary

Key takeaways

News and Commentary

Overview

The video warns that, starting August 2, the next stage of the EU AI Act (Article 50) will trigger mandatory transparency/labeling rules for AI-generated or AI-altered content shared on platforms such as YouTube, Instagram, and TikTok.

The speaker argues this will affect millions of users and companies—not only “high-risk” AI systems—yet it is widely overlooked in public discussion.

What becomes illegal / what must be labeled (from Aug 2)

If AI-generated or AI-altered content appears real, users must be informed that AI was used.

Key points:

  • Labeling must be directly in the content, not only in captions, hashtags, or metadata.
  • Affected content includes:
    • AI-generated or altered images, audio, and video resembling real people/objects/places/events (deepfake-style material, broadly defined).
    • AI-generated or AI-edited “informative texts” on matters of public interest (e.g., politics, health, consumer issues), with important carve-outs (see below).

Penalties (as mentioned in the commentary)

  • Non-compliance can lead to fines up to €15 million
  • Potentially also up to 3% of global annual turnover (whichever is higher, per the commentary)

Who is responsible: providers vs operators

Providers

Companies that place AI systems on the market (e.g., OpenAI, Google, Midjourney).

Operators

Everyone who uses these systems for their own responsibility, including:

  • agencies
  • online shops
  • creators (including influencers/YouTubers making money)
  • companies remain responsible even if work is outsourced

The speaker emphasizes that accountability rests with legal entities (the company/publisher), not individual employees.

Definition and scope of “deepfakes” (broad interpretation)

The video stresses that the EU definition is not limited to celebrity fakes. It covers AI content that noticeably resembles real humans/objects/places/events and could be mistaken as genuine—even if:

  • the depicted person/scene doesn’t literally exist, as long as it looks human/realistic enough
  • there is no intent to deceive

The speaker also highlights gray areas (e.g., AI retouching such as removing objects from backgrounds) and points to newly issued EU guidance dated July 20 (with examples/FAQs), noting it is not legally binding.

Text labeling rules (narrower than some expect)

The speaker clarifies that not every AI email or every ad must be labeled. The obligation applies to:

  • published informative texts on public-interest topics (politics/health/consumer issues)

Excluded (as described):

  • advertising and product descriptions
  • internal communications

Human editorial responsibility exemption (as described)

A key exemption is mentioned:

  • If a human editor/editorial reviewer checks and takes responsibility for the text, additional labeling may not be required.

The speaker warns that:

  • spellchecking or using a second AI is not sufficient
  • human editorial responsibility is required

How labeling must appear (practical placement)

The video argues:

  • Labels must be perceptible without technical aids—at the latest upon first direct contact with the content.
  • A voluntary EU code of conduct suggests symbols such as:
    • “AI generated” for fully AI-made content
    • “AI modified” for AI-edited content

However, the speaker notes:

  • The code is voluntary, so symbols are not strictly mandated—yet improvising can increase legal risk.

Platform/format guidance emphasized

  • Images/videos: label should be embedded within the content itself (e.g., visible in a corner of an image or video), not only in captions.
  • Videos: the AI notice should appear at the beginning and be repeated after interruptions where AI-generated content returns.
  • Audio: requires an audible disclaimer.

The core theme: make the disclosure part of the user’s immediate experience, not an optional add-on.

Additional timeline mentioned

While the main obligations apply from Aug 2, a further requirement about marking AI-generated/edited content in metadata is delayed until Dec 2, 2026 (and has been postponed again).

Enforcement risk in Germany + competition-law angle

The speaker claims enforcement is set to be aggressive:

  • Germany’s implementation makes the Federal Network Agency a central market-surveillance/complaints body.
  • Anyone who sees unlabeled AI content can report it.
  • Competitors can also use German competition law, including cease-and-desist letters, potentially triggering a wave of warnings for companies that don’t comply quickly.

Media industry pushback and open questions

The video reports resistance from sectors such as radio/TV and photography, including concerns like:

  • whether always adding acoustic/visual signals disrupts creative flow
  • photographers questioning where the line is between permitted “tool-like” editing (e.g., Photoshop retouching) and creating a new image requiring labels

The speaker suggests the legal standard may be hard to apply consistently and fears that, without clear mandatory labeling, it may become nearly impossible to distinguish real from fake over time.

Call to action

The speaker recommends:

  • review what your organization/creator workflow produces using AI
  • document who is responsible for editorial checks/approvals—especially for text exemptions
  • use available EU guidance and participate in a planned webinar on Aug 11 for compliance strategy and training/requirements

Presenters / contributors

  • Christian (identified as a Christian lawyer and partner at WBS Legal in Cologne; main speaker)
  • Moritz Gelen (co-presenter at the planned webinar)

Original video