From Berlin with caution: AI and the blurred lines of trade mark use

By Chezanne Haigh | Consultant 

There is a growing sense that courts are beginning to draw lines around artificial intelligence.

The recent decision by the Berlin Regional Court II (Case No. 52 O 62/26 eV) is a good example of that shift. It may not have made headlines in the same way as some other AI cases but it quietly engages with a foundational question in trade mark law:

When does an AI system “use” a brand in a way that creates legal liability?

 

A restrained starting point in Berlin

The case arose in the context of AI-powered search.

A perfume manufacturer brought urgent proceedings against search engine operator, Google, arguing that its AI-generated summaries were infringing its trade marks.

In this case, the court refused to grant an injunction against a search engine operator whose AI-generated summaries referenced a perfume brand alongside alternative products.

The outcome is less surprising than the reasoning.

The court focused on whether the AI provider was using the trade mark in the legal sense. It concluded that it was not. The summaries were understood as aggregations of third-party content rather than the platform’s own commercial communication.

That distinction is doing important work.

It suggests that, at least in Berlin, not every reference to a trade mark in an AI-generated response will amount to actionable conduct. The concept of “use in the course of trade” still serves as a threshold, even in this new technological setting.

 

A subtle but important contrast

The difficulty is that this approach is unlikely to translate seamlessly across jurisdictions.

In the United States, for example, the analysis would still turn on use in commerce and likelihood of confusion. If an AI output appears to position alternatives to a branded product in a way that feels like endorsement or recommendation, that may be enough to shift the analysis.

Even within Europe, the position is already unsettled. Delivered just days earlier to the Berlin decision, another German decision took a very different approach. In a separate case involving AI-generated content, the Regional Court of Munich treated AI-generated summaries as attributable to the provider itself, effectively viewing the output as the platform’s own statement (Case No. 26 O 869/26).

That contrast highlights an emerging divide.

Some courts are inclined to treat AI as infrastructure. Others are beginning to treat it as a speaker.

Where that line is drawn will shape the scope of liability.

 

What about South Africa?

From a South African standpoint, the Berlin reasoning feels familiar, even if the technology is new.

Our law still asks whether there has been use of a mark in the course of trade and that inquiry has always been grounded in context rather than form alone. The question is not simply what the system does but how its output is understood.

If an AI-generated response is seen as a neutral summary of existing material, a court may be reluctant to characterise it as use by the provider.

But the position may shift where the output begins to look less neutral.

If it reads as a recommendation, an endorsement or a structured comparison that positions alternatives in a commercially meaningful way, the argument for liability becomes stronger. At that point, the distinction between aggregation and communication becomes more difficult to maintain.

South African courts have shown, in other contexts, a willingness to look beyond formal classifications and focus on substance. That instinct is likely to carry through here.

 

A line still being drawn

What the Berlin decision ultimately offers is not a definitive answer but a starting point.

It reflects a degree of caution. A reluctance to extend established trade mark concepts too quickly into a space that is still evolving. At the same time, it leaves open the possibility that different facts, or even a different framing of the same facts, could lead to a different result.

This is perhaps unsurprising.

AI systems do not simply display information. They select, arrange and present it in ways that increasingly resemble human communication. As that continues, the legal character of their outputs will become harder to contain within traditional categories.

 

Final thought

The Berlin court has, for now, drawn the line on the side of restraint.

But as AI systems become more assertive in how they present information, that line will come under pressure.

And the question every jurisdiction will need to answer is a simple one, even if the answer is not:

When does an AI system stop showing the market… and start speaking to it?