AI tools write texts, design logos, voice podcasts and shoot videos in fractions of a second. In marketing, in internal communications, in customer service — in 2026 hardly any company still does without AI-based tools. From a copyright perspective this is a double-edged sword: it opens up room to manoeuvre but also creates risks that many companies underestimate.

Austrian copyright law protects distinctive intellectual creations by natural persons (§ 1 UrhG). Content created without a substantial human creative contribution — i.e. produced purely via a prompt and taken over — regularly does not fall under it. This has two consequences:

  • You yourself acquire no exclusive right in pure AI outputs. Competitors may theoretically use the identical output.
  • If your AI-generated content is taken without a licence, you can hardly proceed against it under copyright law.

The solution: enhance AI output through your own human reworking. As soon as enough creative input is added — selection, combination, editorial reworking, design adjustments — the finished work can be perfectly capable of copyright protection.

The real risk: training data

AI models were trained on gigantic amounts of copyright-protected works — images, texts, code, music. If a model produces outputs that closely resemble a specific protected work, claims by the original rights holders can arise — also against you as the AI user.

First lawsuits against AI providers are pending worldwide. The legal situation is dynamic — and the risk centre is partly shifting to AI users:

  • Images that visibly imitate a well-known trade mark
  • Texts containing literal passages from protected works
  • Code that reproduces proprietary source code
  • Voice clones that infringe personality rights

A cease-and-desist letter can also arrive when you thought you had “just used an AI tool”. Liability follows the output.

EU AI Act: what already applies in 2026

The AI Regulation (EU 2024/1689 — the AI Act) has been in force since August 2024, with its duties phased in until 2027. The following are already relevant in 2026:

  • Prohibited AI practices (Art 5): applicable since February 2025 — e.g. social scoring, manipulative systems
  • Transparency for GPAI models (Art 53): providers of general-purpose AI models have had to publish a detailed summary of their training data since August 2025. This makes it visible to rights holders whether their works were used.
  • Labelling duties for deepfakes and AI-generated content (Art 50): partly already applicable, fully from August 2026

Relevant for end-users of AI tools: anyone distributing AI-generated content that misleadingly depicts persons, places or events must label it. This also applies to marketing material that uses realistic images or videos.

What businesses should do now

From our advisory practice, a 5-point plan emerges:

  • Review the terms of use of all AI tools used for commercial usability — and document the result
  • Check AI outputs before publication for similarity to known protected works (reverse image search, plagiarism tools)
  • Create an internal AI policy: which tools may be used, for which purposes, who reviews the output
  • Build your own creative contribution into the AI workflow — so that outputs become protectable as reworked works
  • Anchor labelling duties in the marketing workflow (especially for realistic depictions of persons)

Those who structure early avoid double costs: first, the risk of a cease-and-desist over training-data conflicts. Second, the damage when AI-generated marketing content is unprotected and copied by competitors.

Additional areas of tension

Beyond classic copyright, the use of AI touches further areas — which we always consider in advice:

  • Unfair-competition law (UWG): AI-generated content without labelling can in certain circumstances count as misleading advertising
  • Personality rights: voice clones, photo manipulations, deepfakes
  • Data protection (GDPR): entering personal data into AI prompts
  • Liability questions: who is liable for incorrect advice by in-house chatbots?

A legal position assessment for AI use typically takes 1–2 hours — and creates clarity about what you must already observe today, where there’s still room to manoeuvre and where you should already be preparing for the next regulatory wave.

This article is for general information and does not constitute legal advice in any specific case.