EU Objectives

As part of its digital strategy, the EU aims to regulate artificial intelligence in order to create better conditions for the development and use of this innovative technology. AI can bring many benefits, such as better healthcare, safer and cleaner transportation, more efficient manufacturing, and a cheaper and more sustainable energy supply.

The European Parliament aims, in particular, to ensure that AI systems used in the EU are safe, transparent, traceable, non-discriminatory, and environmentally friendly. AI systems should be monitored by humans rather than through automation in order to prevent harmful outcomes.

Citizens have the right to file a complaint with the relevant national authorities regarding the use of AI systems.

Risk is the key factor

Most AI systems pose only a low risk, but they must nonetheless be assessed for risk. Based on the determined risk, certain obligations then apply to AI users as well.

There are three risk levels:

Low Risk / Transparency Requirement

The use of AI in day-to-day editorial work is usually classified as “low risk.” AI systems classified as low-risk must meet transparency requirements. AI systems such as ChatGPT, for example, fall into this category.

The EU distinguishes between the following:

• Simpler generative foundation models such as ChatGPT (presumably referring to the GPT-3 version) and
• General-purpose AI systems with significant impact that could pose a systemic risk, such as the advanced AI model GPT-4.

With AI systems at this risk level, users can create images, videos, audio files, and text.

Labeling Requirements

To meet transparency requirements, AI-generated content must be clearly labeled as “AI-generated” under certain conditions. Users should be able to recognize that content—such as images, audio, or video files—has been created or edited using AI.

This applies in particular to “deepfakes.” Deepfakes must be clearly labeled. “The relevant technical solutions must be effective, interoperable, robust, and reliable, and must take into account the specific characteristics and limitations of the various types of content, the implementation costs, and the generally accepted state of the art, as may be reflected in the relevant technical standards. (Quote from the European Commission)”

Texts that have been created or edited using AI must be labeled as such if they are intended to inform the public about matters of public interest.

As a general rule, text does not need to be marked if

• the texts were reviewed through a human verification process (e.g., during AI-based profiling in the context of police investigations) or through editorial review, or
• A natural person or legal entity that assumes editorial responsibility for the publication.

Copyright and Rights of Public Performance

In this context, it is important to distinguish between an AI's input and output.

AI output

AI-generated text and images are not protected by copyright because they are not intellectual creations that can be attributed to specific individuals. A prompt created by a human sets a goal, but the AI does the creative work.

AI-generated content may be used freely as long as the content itself does not infringe on any rights. Anyone may copy and use AI-generated content, for example, for their own website or for their business.

AI-generated content may infringe on rights; for example:

• If, by chance, unmodified training data was output.
• When copyrighted content is displayed, such as logos, trademarks, or characters from movies or comics.
• When personal rights are violated.
• When a style—such as a writing style or a painting style—is to be imitated, resulting in something similar to the original.

It is rare, but possible, for an AI to happen to output training data that has not been altered. The training data has been incorporated into the mathematical structure of the neural network but has not been stored in its original form. Manufacturers are permitted to use publicly available texts to train AIs, provided that the texts are not permanently stored.

Input into an AI

The user enters data into an AI using what are known as prompts. A prompt is a set of instructions for an AI.

A prompt may be protected by copyright if it reaches a “necessary level of creativity.” This could be, for example, an original story intended to serve as the basis for creating an image. In that case, the prompt is protected by copyright. The AI-generated image, however, remains unprotected and may be used freely.

For example:

Privacy and Personal Information

For safety’s sake, any publicly accessible AI should always be viewed as a black box that stores all input data. Even if the AI itself does not store anything, there is no way of knowing what the AI provider stores. AI users have no control over what happens to the data.

The General Data Protection Regulation (GDPR) applies. No personal data—such as that of customers, partners, or colleagues—may be entered into an AI system without explicit consent.

Rights of Personality / Right to One's Own Image

Data relating to individuals or containing other protected content may not be uploaded without express consent; for example, images of people or audio files containing voices.

Conclusion – What Technical Writers Need to Keep in Mind

As a general rule, the content of AI-generated text and images should always be reviewed before use and should not be shared with third parties without first being reviewed. At this time, we strongly advise against publishing such content without reviewing it first.

A liability issue may arise if the above rules are not followed. Damage may occur if AI output is disclosed without proper oversight.

Before personal data or other sensitive or protected data may be used in a prompt, the affected individuals or legal entities must give their consent.

AI-generated text and images are not protected by copyright and may be used freely, as long as they do not infringe on any rights or violate any laws.

In a newsroom, AI can be considered a “normal” work tool if the AI’s output is reviewed as part of a human review process or through editorial oversight. In that case, the generated texts do not need to be labeled.

The EU AI Regulation will take effect in several phases

The law titled “Regulation (EU) 2024 establishing harmonized rules for artificial intelligence” was adopted by the Council in May 2024.

Starting in November 2024
The EU AI Regulation will take effect.

Effective February 2025
AI systems that pose unacceptable risks are prohibited. Prohibitions, definitions, and regulations related to AI competencies apply.

Effective August 2025
The transparency requirement for the use of AI applies.
The provisions on governance (rules for national legislators) and the obligations regarding general-purpose AI apply.

Effective February 2026
Following consultation with the European Artificial Intelligence Board, the Commission will provide guidelines on practical implementation and a comprehensive list of practical examples of use cases for AI systems that are either high-risk or non-high-risk.

Starting in August 2026

The law will take full effect.
Different deadlines apply to manufacturers of high-risk AI systems.

Key Sources:

European Commission: “Artificial Intelligence – Questions and Answers” – As of August 1, 2024
https://ec.europa.eu/commission/presscorner/detail/de/qanda_21_1683

European Parliament: “AI Regulation: First Regulation on Artificial Intelligence” – As of June 19, 2024
https://www.europarl.europa.eu/topics/de/article/20230601STO93804/KI-Verordnung-erste-regulierung-der-kunstlichen-intelligenz

REGULATION (EU) 2024 / OF THE EUROPEAN PARLIAMENT AND OF THE COUNCIL establishing harmonized rules on artificial intelligence
https://data.consilium.europa.eu/doc/document/PE-24-2024-INIT/de/pdf