AI and copyright – what you need to know

The focus is on what is known as the ‘degree of originality’. A work is only protected by copyright if it reflects an original creative contribution, in other words, a ‘personal intellectual creation’.

This is precisely where the challenge with AI lies:
Purely automatically generated content does not, as a rule, fulfil this requirement. Although it is created on the basis of prompts, the actual design lies in the way the AI functions, and not in a specific, controllable creative decision made by a human being.

‘With regard to AI, therefore, copyright law only applies if the basis for the newly created work was originally created by a human being.’ (Source: Federal Ministry of Justice)

In practice, this means:
An AI-generated text or image does not, at first, have a traditional author. A simple prompt is therefore rarely sufficient. A carefully crafted text, a customised image or a video is more likely to qualify.

Things get exciting when humans come back into the picture. As soon as an AI-generated result is specifically refined, edited or creatively developed, it can become a work in its own right. The crucial factor is that human input is the defining element.

This means:
AI often provides the rough framework or a relevant impetus. The part that is relevant for copyright purposes only comes about through selection, adaptation and creative refinement. So the more we control, shape and add creative finishing touches, the more likely it is to become our own work.

Here is a brief overview of scenarios for guidance:

1. Purely AI-generated (prompt only → result)

• no or very little human influence
• The result is predominantly created by the system

We generally have NO copyright protection. The problem:

• In theory, anyone could use or copy the image

2. AI + simple editing

• minor adjustments
• light retouching / selection

This is usually still problematic / no clear protection. The level of creativity is often insufficient.

3. AI + significant creative editing

• Composing
• own creative concept
• Combination of several elements
• distinctive creative style

This may give rise to copyright, meaning that you then become the author of the work as a whole

A company uses AI to generate a logo. The result looks good and is adopted straight away.

The problem:

• The logo is generated entirely by AI → no clear copyright protection
• similar versions could appear again at any time
• in the worst-case scenario, it is too similar to existing logos

The far better approach (agency practice):

• AI provides initial directions and style ideas
• Designers then develop their own concept based on this
• Shapes, typography and the figurative mark are customised
• The logo is deliberately simplified, refined and made brand-ready

Result:

• unique
• recognisable
• legally much more robust

Conclusion: What matters is not that AI was used, but rather the extent to which human creativity has contributed to the result.

A logo is not just any old piece of content, but the visual foundation of a brand. If it is interchangeable, it lacks clear recognisability and, consequently, part of its brand value. At the same time, the legal risk increases, as generic or AI-generated logos are often difficult to protect and may be too similar to existing brands.

It is important to make the following distinction:

a) Training AI – The crucial question is: where does the data come from?

Many large AI models have been (and are being) trained using vast amounts of data – presumably including copyright-protected content.

And this is precisely where it gets interesting:

• Authors, artists and publishers have already joined forces
• legal proceedings are underway against major AI providers
• in some cases, it has been shown that AI reproduces content that is very similar to original works

In principle, so-called text and data mining is possible, but only if rights holders have not objected to it.

Liability: Who is responsible?

The issue of liability has not yet been conclusively clarified, but from the user’s perspective, the following applies: if you publish content,you are initially liable.

This means:
Even if the AI has generated something, you may still be held liable for any copyright infringements. For example, if:

• works by others have been recognisably reproduced
• the content is too closely modelled on existing works
• protected content is reproduced

The problem, therefore, is not just whether you hold the copyright, but whether you might unwittingly infringe someone else’s rights.

b) AI output (what is ultimately used?)

When using AI outputs, however, the question arises as to whether protected works are still recognisable within them. If this is the case, the rights of third parties may be affected. If they are no longer recognisable, the use is in many cases less problematic.

This question is, of course, becoming increasingly relevant. We put a lot of effort into our content, then someone comes along and brazenly copies it all into, say, ChatGPT… And I find myself wondering:
“How do I prevent my content from being used?”This can be donevia a so-called ‘opt-out’ clause:

• Content can be marked in such a way that it must not be used for AI training
• Important: This must be done in a machine-readable format

Only then can systems actually recognise and take this into account. But what I find even more important is: create content that cannot simply be reproduced.

• Your own data
• Your own experiences
• clear opinions
• Your own distinctive style 

Even though AI automates many processes, the existing rules remain in place. This applies not only to copyright but also to related areas such as data protection and personal rights. The situation can quickly become sensitive, particularly when it comes to images, videos or the voices of real people.

It is important to note that as soon as personal data is involved (e.g. in images, videos or texts relating to individuals), the established rules on consent, purpose limitation and data security apply.

The transparency requirement is coming (and makes sense)

The labelling of AI-generated content is becoming increasingly important under the EU AI Regulation – particularly in the case of deceptively realistic content such as deepfakes.

This article deliberately focuses on copyright law.
We have set out the details of the labelling requirement – when it applies and how it is to be implemented in practice – separately here .

AI is fundamentally changing content production – but not the foundations of copyright law.

The crucial point remains: it is not the ‘machine’ that creates the protected work, but the human being. Anyone using AI should always ask themselves:
‘How much of my own creative input is actually in the final result?’

Never publish ‘blindly’. Ask yourself:

1. Where does the data come from?
2. How much of your own work is in it?
3. Is it clear to users that AI was involved?

In addition, pay attention to:

1. data protection guidelines for personal data
2. the labelling requirements for AI-generated content that is deceptively realistic or sensitive

The legal landscape surrounding AI is currently evolving at an extremely rapid pace, particularly at EU level. New rules, greater clarity, but also more obligations are on the horizon.

Stay up to date to avoid risks, but above all to deliver really strong content. We are currently entering a phase in which:

• ‘producing content’ is no longer anything special
• ‘Having original ideas’ is becoming the real asset

And that’s actually a positive development. Because, in my view, ultimately it’s those who use AI to create something entirely their own who will come out on top.