Artificial Intelligence Enters the Courtroom (But Stays Out of Evidence): The Case of the Turin Court of Appeal

Artificial Intelligence Enters the Courtroom (But Stays Out of Evidence): The Case of the Turin Court of Appeal

In the field of Intellectual Property, we are accustomed to seeing world-renowned experts clash over minute technical details. However, recent judgment No. 570/2026 of the Turin Court of Appeal, published on 19 March 2026, has introduced a new technological “third party” into the debate on patent validity: Artificial Intelligence (AI).

This case offers a fundamental lesson on the legal limits of tools such as ChatGPT and search engines when used as “expert witnesses.”

The Heart of the Dispute: Between Strength and Lightness

The dispute concerned the validity of the Italian portion of European patent EP 1 951 483. The invention describes a cutting machine with an outwardly bulged depression box.

The technical advantage? A combination as simple as it is difficult to be achieved: increasing structural resistance to pressure (necessary for cutting) while simultaneously reducing the machine’s weight by 30%.

The appellant argued that this solution was “obvious.” To demonstrate this, in addition to citing prior art from distant sectors such as insulated containers, they decided to play the modernity card: they queried ChatGPT and Google, claiming that these tools—by scanning global knowledge—confirmed that the principle of curved walls to withstand pressure has been known for centuries (as in submarines or dams).

The Court’s Position: Why AI Does Not Constitute Evidence

The Turin Court of Appeal firmly rejected the use of AI-generated responses as technical evidence of the obviousness of an invention. The judges’ reasoning touched on key issues for the future of IP litigation:

• Procedural inadmissibility: AI queries submitted late in the proceedings were declared inadmissible, effectively constituting “new documents.”
• Lack of reliability: The Court noted that no proof was provided regarding the robustness of the applications used. The scope and quality of the datasets on which the AI was trained are unknown, as is its capacity for technical discernment.
• The risk of “hallucinations”: The judges expressed concrete concern about the ability of such applications to avoid so-called “hallucinations,” i.e., the generation of technically incorrect answers presented in an authoritative tone.
• Statistics vs. technique: The ruling emphasised that AI does not “invent” or “understand,” but provides responses based on the statistical probability of word sequences. This process is worlds apart from the rigor required to assess inventive step in relation to the state of the art.

The key point: “No evidentiary value can be attributed to the answers produced by the appellant” through AI, since the quality and depth of the underlying data cannot be verified.

An Invention Is Not Just a Physical Principle

The appellant attempted to reduce the invention to a trivial application of a basic physical law. However, the Court clarified that patentability lies precisely in the original technical implementation of those principles to solve a specific problem in a non-obvious way for a person skilled in the art.

Using a striking comparison, the judges observed that if the argument of obviousness based on known physical laws were accepted, we would have to invalidate all patents on smartphone accelerometers simply because they are based on the “well-known law of gravity.”

Avoiding “Hindsight Bias”

Another pillar of the judgment concerns the so-called Problem-Solution Approach. The appellant had attempted to combine a patent for a flat cutting machine with that of a double-walled insulated container (document D2).

However, the Court, following the conclusions of the court-appointed technical expert (CTU), established that:

1. The sectors are technically distant (textile machinery vs. thermal insulation).
2. Extracting only the “bulging” from document D2, while ignoring its complex double-wall structure, amounts to an ex post facto (“hindsight bias”) analysis. In other words, it is easy to claim something was obvious after already seeing the patented solution.

Conclusions for Companies

This judgment represents an important “stop” to attempts to oversimplify patent validity assessments through generative AI tools.

For companies, the message is clear:

• AI is a support tool, not a substitute for technical and legal expertise.
• Patent validity must be defended and challenged through rigorous analysis of the state of the art and proper application of legal criteria (novelty and inventive step), not through chatbot screenshots.
• The quality of technical expertise (both court-appointed and party-appointed) remains the only real basis for judicial assessment.

In a world racing toward automation, the Turin Court of Appeal reminds us that judgment on human ingenuity still firmly belongs to humans and the law.

Our firm remains at our Clients’ disposal to assist them in analyzing specific cases, helping them protect their inventions and defend their patent rights.