Artificial intelligence is rapidly entering everyday legal practice: from case law research to drafting briefs, to litigation strategy. But what happens when AI-generated content is used as evidentiary material in court?
A recent order from the Court of Ferrara dated February 20, 2026 offers a clear answer: a conversation with ChatGPT is not an evidentiary document and cannot be used to prove the merits of a legal claim.
The decision represents one of the first Italian cases in which a judge directly addresses the procedural use of responses generated by artificial intelligence systems, placing the issue within the emerging European regulatory framework outlined by the AI Act and the GDPR.
The case: a road accident and "digital" evidence
The proceeding arose from an application under Article 696-bis of the Italian Code of Civil Procedure, seeking a preliminary technical assessment to reconstruct the dynamics of a fatal road accident.
The applicant argued that the highway concession company was also liable due to inadequate signage and traffic management.
Among the documents filed was a particular item:
A conversation with ChatGPT citing case law precedents in support of the defense.
The court analyzed the document and reached a very clear conclusion:
- •the document was partial, as the original prompt was missing
- •the cited rulings were not relevant to the case
- •generative AI systems are subject to hallucinations — producing plausible but inaccurate information
Consequently, the judge stated that a chatbot conversation:
Is not a procedural document
Does not constitute atypical evidence
Must be considered tamquam non esset
i.e., as if it did not exist
AI and "hallucinations": the reliability problem
The central point of the decision concerns the probabilistic nature of generative AI.
These systems:
- •do not necessarily retrieve verified data
- •generate text based on linguistic probabilities
- •can produce non-existent legal citations or irrelevant ones
The court explicitly references the notion of AI hallucinations, i.e., situations where the system:
- •invents legal references
- •confirms them in subsequent queries
- •thus produces an apparent informational authority
For civil proceedings, where evidence must be verifiable and traceable, this creates a structural problem.
The role of AI under the European AI Act
The order expressly links the issue to the AI Act (EU Regulation 2024/1689).
The European regulation introduces two fundamental principles:
1. Human oversight
AI systems must be used under meaningful human supervision.
- •the professional user must verify results
- •cannot delegate legal decisions or assessments to AI without oversight
2. Responsible AI
AI use requires a responsible and verifiable approach.
- •generated information must be source-checked
- •cannot be automatically used as the basis for argumentation
The judge emphasizes that these principles now have direct relevance in Italian law through national transposition.
AI and professional liability of lawyers
One of the most interesting aspects of the order concerns the liability of professionals who use artificial intelligence.
According to the court:
- •chatbots are support tools
- •they cannot replace the lawyer's professional oversight
- •source verification is a duty, not just advisable
In practice, the risk is twofold:
Procedural risk
Briefs based on inaccurate sources may be disregarded or sanctioned.
Ethical risk
The lawyer remains responsible for the content presented in court.
In recent years, several Italian courts have already flagged cases of improper AI use in drafting legal briefs.
The GDPR dimension
The decision also references the General Data Protection Regulation (GDPR) framework.
The interaction between generative AI and personal data raises at least three legal issues:
1. Legal basis for processing
When a professional inputs client data into an AI system, this constitutes personal data processing and requires a valid legal basis.
2. Data minimization
The data minimization principle prohibits sharing sensitive data or unnecessary information with external systems.
3. Professional liability
The lawyer remains the data controller or processor and must assess where data is processed and how it is stored.
Indiscriminate use of AI tools in legal practice can therefore create significant GDPR risks.
AI as a tool, not a source
The main lesson from the Court of Ferrara's decision is simple but fundamental.
Artificial intelligence can be useful for:
- •preliminary research
- •legal brainstorming
- •document summarization
But it cannot replace:
- •source verification
- •legal reasoning
- •professional responsibility
Chatbots remain tools at the service of people and cannot become autonomous sources of evidence in civil proceedings.
A precedent set to make history
This order could become an important reference in Italian case law on the use of AI in the legal field.
Because it clarifies three crucial points:
- •chatbot responses are not evidence
- •AI use requires human verification
- •professionals remain responsible for the content produced
In court, technology can assist the law — but it cannot replace it.
Does your organization use AI in legal practice?
We can help you define responsible AI usage policies, in compliance with the AI Act and GDPR.
Request a consultation