ChatGPT and professional secrecy, what professionals risk
In Italy «legge 132 2025», the Italian law on artificial intelligence, is searched on Google about 1,900 times a month, «segreto professionale» (professional secrecy) about a thousand, «chatgpt avvocato» (ChatGPT lawyer) 880 and «art 622 codice penale» (article 622 of the criminal code) 590. These are Google Keyword Planner figures for the last year, and behind them I picture very concrete questions. A lawyer who would like a brief summarised, an accountant who wants a draft reply to a client, a psychologist thinking of tidying up the notes from a session.
AI can genuinely help in all these cases. The problem is that, to do so, you almost always hand it a text containing information covered by professional secrecy. In this article I try to bring together what the rules say, what happens to the data and which habits let you use AI without betraying the trust of the people who come to you.
Secrecy does not change with AI
The starting point is the Italian criminal code. Article 622 punishes anyone who, having learned a secret through their profession, discloses it without just cause or uses it for their own or someone else's profit, if harm may result. The penalty is imprisonment of up to one year or a fine, and the offence is prosecuted only on the complaint of the injured party.
Alongside the law there are the codes of conduct, which for each profession repeat the same principle in different words.

Professional secrecy and artificial intelligence in Italian codes of conduct and law. Source, official texts of CNF, CNDCEC, FNOMCeO, CNOP and Normattiva, checked on 24 October 2026.
For lawyers, the code of conduct of the Italian National Bar Council (CNF) calls secrecy a duty, as well as a right, that is «primary and fundamental», and requires lawyers to have it respected by employees, trainees, consultants and collaborators too, even occasional ones. For doctors, the code of medical ethics requires secrecy on everything learned through the profession. For psychologists, the code of conduct is even firmer, because the psychologist is «strictly bound» to secrecy and may waive it only with the person's valid and provable consent.
Accountants are the only ones who already have specific ethical rules on AI. With the amendments in force since 21 November 2025, the code of their National Council (CNDCEC) states that AI may be used only for instrumental and support activities, that professionals must make sure the systems have adequate security and confidentiality measures and comply with data protection rules, and that using AI is never an excuse for violations. For doctors, the national federation (FNOMCeO) started work on a new code in April 2026, aiming to approve it in 2027, and AI is among the topics.
What the AI law requires
Italian law 132 of 2025, in force since 10 October 2025, devotes article 13 to the intellectual professions. It says two things. AI may be used only for instrumental and support activities, with the professional's intellectual work prevailing. And information about the AI systems used must be given to the client «in clear, simple and comprehensive language». On 13 October 2025 the National Bar Council sent the bar associations an information template that lawyers can use, if they find it helpful.
On 23 October 2026 Legislative Decree 179 of 2026 came into force, adding two new elements for professionals.

The main milestones. Source, Gazzetta Ufficiale, Normattiva, Garante privacy, CNF, CNDCEC and OpenAI, checked on 24 October 2026.
Article 47 requires the initial and continuing training of the professional orders to include courses on AI, covering technical, legal and ethical aspects, including the duty to inform the client. The national councils must adapt their regulations within six months. Article 49 provides that the use of AI may lead to an adjustment of fair remuneration, according to parameters still to be defined and linked to the risk class of the system used.
What happens to data in ChatGPT
Here the type of account matters a lot. On ChatGPT's personal plans (Free, Plus and Pro) OpenAI writes that sharing data to improve its models is on by default. You turn it off with the «Improve the model for everyone» setting, but there is a detail many people do not know. If, after turning it off, you leave feedback, for example with a thumbs up or down, the entire conversation linked to it may be used for training. Temporary chats are not used to improve the models, but OpenAI may keep them for up to 30 days for safety purposes.
On business plans, such as ChatGPT Business and Enterprise, OpenAI writes that by default it does not use business data to train its models, and it offers a data processing agreement (DPA) to sign. For a time, moreover, a US court order in the lawsuit brought by the New York Times required OpenAI to retain users' chats, including deleted ones. OpenAI announced that this obligation ended on 26 September 2025.
In July 2025 the Italian data protection authority (Garante privacy) raised the alarm about medical reports uploaded to chatbots, speaking of a risk of losing control over very sensitive health data, and stressed the need for qualified human involvement. The message applies to anyone who handles client or patient data.
Five habits for the practice
From all this I draw five practical habits, valid for any professional practice.
- Never identifying data in a personal chatbot. Names, tax codes, case numbers, diagnoses and details that make a person recognisable should not be pasted into a free or personal account.
- Replace the data before, not after. If you really need to work on a document, remove the personal data first. You can do it by hand, as I explain in the guide on how to anonymise texts for chatbots, or with the Anonimizzatore IA (AI Anonymiser) that we distribute free of charge for macOS and Windows. It replaces personal data with reversible placeholders, your documents stay on your computer, and afterwards you can put the real names back into the answer. The result must always be checked, because no tool recognises everything.
- If your practice uses AI, choose a business plan with a DPA, check the settings and keep track of what you use.
- Inform the client, as article 13 of law 132/2025 requires, in plain words. The CNF template can be a good starting point for other professions too.
- Train yourself, and train your collaborators. Secrecy applies to them too, and decree 179/2026 brings AI into the mandatory training of the professional orders.
If you want to receive updates like this one when the rules change, create your free account on retoria.it (the form is in Italian). If instead you want a course on using AI for your practice or your professional order, with real cases and attention to secrecy and data, write to us.
Methodological note
On 24 October 2026 I read the official texts of law 132/2025 and Legislative Decree 179/2026 (Gazzetta Ufficiale and Normattiva), article 622 of the criminal code, the codes of conduct of lawyers (text updated to 2026), accountants (with the 2025 amendments), doctors and psychologists, the CNF's notice to the bar associations, the FNOMCeO page on the new code, the Garante's press release on medical reports and OpenAI's official pages on data controls, business privacy and the New York Times lawsuit. Some OpenAI pages block automated reading, so I consulted them in the browser.
I do not report the amount of the fine under article 622, because the text on Normattiva still states it in lire. The lawyers' code of conduct, in the text updated to 2026, does not yet contain specific rules on AI. Search volumes come from Google Keyword Planner for Italy, from October 2025 to September 2026. This article is not legal advice, and for specific cases it is best to contact your own professional order or a lawyer.
Sources
- Law 23 September 2025 no. 132, article 13 on Normattiva and text in the Gazzetta Ufficiale (GU no. 223 of 25 September 2025).
- Legislative Decree 7 October 2026 no. 179, article 47 and article 49 on Normattiva (GU no. 234 of 8 October 2026, in force since 23 October 2026).
- Italian criminal code, article 622.
- National Bar Council (Consiglio nazionale forense), code of conduct updated to 2026, article 28, and notice to the bar associations on the AI information template (13 October 2025).
- National Council of Chartered Accountants (CNDCEC), code of conduct with the 2025 amendments, articles 10 and 21.
- FNOMCeO, code of medical ethics, article 10, and press release on the ethics council (29 April 2026).
- National Council of Psychologists (CNOP), code of conduct of Italian psychologists, articles 11 and 12.
- Italian data protection authority (Garante per la protezione dei dati personali), press release on medical reports and artificial intelligence (July 2025) and press release on blocking ChatGPT (March 2023).
- OpenAI, data controls in ChatGPT, enterprise privacy and response to the New York Times' data demands (updated 22 October 2025).
- Google, Keyword Planner, average monthly search volumes in Italy (October 2025, September 2026), checked on 24 October 2026.