Insights
7 mistakes lawyers make with ChatGPT (and the court rulings that sanctioned them)
3 min read · Updated March 2026 · Editorial oversight: Avv. Federico Papa
ChatGPT entered Italian law firms before a clear regulatory framework was established. Italian courts have already sanctioned counsel for non-existent citations and uncritical reliance on AI-generated briefs: notably, the Courts of Florence (2025), Turin (2025), and Syracuse (2026), with penalties exceeding 30,000 euros. This guide outlines the 7 documented mistakes and how to avoid them in practice.
In brief
The Courts of Florence, Turin, and Syracuse sanctioned the uncritical use of AI in legal practice between 2025 and 2026. Penalties exceeding 30,000 euros addressed the inclusion of non-existent Supreme Court citations and violations of Art. 96 of the Code of Civil Procedure regarding abuse of process. Documented errors involve processing confidential data through consumer interfaces, lack of critical legal review, and using generative models as statutory databases. Professional responsibility and authorship remain with the signing lawyer, requiring mandatory verification via official sources.
- 1.
Trusting unverified case-law citations
This mistake lies at the root of the first sanctions. In the first Italian decision (Court of Florence, March 2025), ChatGPT fabricated Supreme Court judgment numbers in a trademark dispute and, upon further prompting, re-confirmed their authenticity. In Syracuse (February 2026), relying on four non-existent judgments resulted in costs and penalties exceeding 30,000 euros, including legal fees, damages for vexatious litigation, and fines: the court ruled that awareness of model hallucinations is now common knowledge expected of every practitioner.
- 2.
Filing briefs without rigorous legal review
The Court of Turin (September 2025) described a petition drafted entirely with AI as an abstract collection of statutory and case-law references, lacking logical structure and largely irrelevant, sanctioning counsel for abuse of process under Art. 96 of the Code of Civil Procedure. AI-generated text must never be filed as-is; it must always be restructured according to sound legal logic.
- 3.
Pasting data covered by professional secrecy
Client names, factual background, and confidential documents pasted into a consumer chat interface may be retained by the provider and, depending on settings, used to train models. Prior to any professional use, it is essential to check the data retention and training settings of your OpenAI account and to systematically anonymise all case-file information.
- Try edit.legal
Apply this research directly with edit.legal
Legal research and drafting with citations checked against official databases. Try edit.legal for free, no credit card.
4.Using the tool as an up-to-date statutory database
The model answers based on the training dataset: recent legislative reforms, corrective decrees, and repeals may be missing or misrepresented. The only reliable references remain official consolidated sources (such as Normattiva or the Official Gazette) or specialised tools capable of citing and verifying sources directly.
- 5.
Asking generic questions and accepting abstract answers
A prompt lacking factual details, dates, and precise legal qualifications yields a textbook summary rather than a case-specific analysis. Since the model does not request missing details, the onus is on the practitioner to provide a comprehensive factual context and ensure that the response directly addresses the concrete case.
- 6.
Ignoring professional duty and ethical obligations
Using AI does not shift professional responsibility: ownership of the brief remains with the signing lawyer, together with the duties of competence, diligence, and candour toward the court. Case law from 2025 and 2026 classifies the failure to verify generated citations as gross negligence: blind delegation to the tool is itself a breach of professional duties.
- 7.
Using the wrong tool for legal work
Related practical guide: How to draft a reasoned legal opinion with AIChatGPT excels at rephrasing, summarising user-provided texts, and drafting non-technical outlines. Conversely, legal research, citation retrieval, and brief drafting require professional tools with verified Italian sources and automated cross-checks against official databases: this represents the critical distinction between a general writing assistant and a dedicated legal practice tool.
References: Trib. Firenze, marzo 2025Trib. Torino, settembre 2025Trib. Siracusa, febbraio 2026art. 96 c.p.c.
Frequently asked questions
Can I use ChatGPT to write a court petition?
You may use it for initial drafting or rephrasing, but not for legal research or citation retrieval: Italian courts have already sanctioned reliance on unchecked AI-generated petitions. Every statutory and case-law reference must be verified against official sources prior to filing.
Is ChatGPT GDPR-compliant regarding client data?
Compliance depends on the subscription plan and settings configured: consumer versions may retain conversations and use them for model training unless expressly opted out. When processing data covered by professional secrecy, adequate contractual guarantees or rigorous prior anonymisation of all identifying details are required.
What sanctions does a lawyer risk for AI-invented citations?
Decided cases include adverse costs orders, aggravated liability under Art. 96 of the Code of Civil Procedure for vexatious litigation, monetary sanctions (exceeding 30,000 euros in the Syracuse judgment), and potential disciplinary proceedings before the Bar Council for breach of the professional duties of diligence and candour.
Verified legal research and drafting with edit.legal
Legal research and drafting with citations checked against official databases. edit.legal is free to try, no credit card.
Try edit.legal for free