Chartered accountants and artificial intelligence, inside and outside the firm

The commercialisti (chartered accountants) registered on the professional roll numbered 119,050 at the end of 2025, working in about 69,000 firms. More than half of the firms have three people or fewer. Their archives hold almost everything confidential that a client has: income, bank accounts, contracts, payslips, and, for anti-money-laundering purposes, identity documents and beneficial owners, kept for ten years.

Artificial intelligence has already entered these firms. According to the 2025 survey by the Fondazione Nazionale dei Commercialisti (the profession's national foundation), covering over 4,000 professionals, 34% use it a lot or fairly often and more than 80% have tried it at least once. The Politecnico di Milano Observatory finds that in 2026 39% of accountancy firms use artificial intelligence tools outside their management software. The most cited uses are searching rules and documents, recording electronic invoices, bank reconciliation, and drafting texts and letters.

What the rules say

The law. Since 10 October 2025, Law 132 of 2025, in Article 13, allows professionals to use artificial intelligence only for instrumental and support activities, and obliges them to tell the client which systems they use. It does not forbid the cloud and does not require on-premises systems.

The code of ethics. On 20 November 2025 the Consiglio nazionale dei dottori commercialisti e degli esperti contabili (the national council of the profession) added three paragraphs to Article 21 of the code of ethics, in force from the following day. A professional who uses artificial intelligence "takes full responsibility and control for it" and must, among other things, "make sure that the artificial intelligence systems used have adequate security and confidentiality measures and comply with the rules on personal data protection". The professional must also inform the client of the systems used, "also in any documentation produced". The obligation in the law has become a disciplinary obligation as well, and checking the security of the tool is a personal duty.

Secrecy and personal data. Professional secrecy (Article 5 of Legislative Decree 139 of 2005), the confidentiality rule of the code of ethics and the European data protection regulation all remain: if an external service receives clients' personal data, a contract appointing it as data processor is required.

What the professional bodies say

The third operational guide by the National Council and the Foundation, presented at the Genoa congress in October 2025, is clear about free tools for the public: their terms usually allow content to be used for training, and "providing such a service with a document containing a client's sensitive information would amount to a breach of professional secrecy and of the GDPR". As a contractual answer it points to the business versions from OpenAI, Google, Microsoft and Anthropic, with their guarantees.

The same guide contains a seventeen-page appendix on installing a language model locally, presented as "a concrete and accessible strategic choice for any professional firm that considers data privacy and technological autonomy as priorities". The limit it names is maintenance, which requires "a certain level of technical competence that may not be immediately available within a professional firm".

The guide of the Turin Order (the local chartered accountants' body), from June 2026, distinguishes three cases. Clients' personal data on a free platform: forbidden, a breach of the European regulation and of the code of ethics. Business platform: allowed, with a data processing contract, known servers, no training on the data and a written notice to the client. Model run locally: no external transmission, with the security of workstations and access still to be looked after; the guide points to it where confidentiality is "stringent", as "a technically more burdensome option, but in certain contexts decidedly appropriate".

The rules exist; their application lags behind. In the data presented at the 2026 national conference of young commercialisti, fewer than 5% of firms have a written policy on artificial intelligence, fewer than 3% say they know the rules well, and about a quarter began using it before having internal rules.

What the management software offers

For many firms the shortest route is the artificial intelligence inside the management software they already use. TeamSystem Studio has an assistant that answers in natural language, proposes accounts and records invoices and bank statements; Zucchetti Ago Infinity has the assistant Aladino and predictive models for postings; Wolters Kluwer has brought artificial intelligence into tax returns with Genya; Sistemi offers it for PROFIS. All of these are cloud services; where the models run and who the providers are, the public pages do not say, and prices are not published.

It has a real advantage, which the Turin guide recognises: the data stay with a provider with whom a contract already exists, and the assessment of data processing is simpler. It also has the limit of every centralised archive. In August 2026 TeamSystem confirmed the exfiltration of data from its Contabilità in Cloud service, used by small businesses and accountants: personal details, IBANs and accounting entries. No solution removes the risk; it changes where it sits and how many clients it touches at once.

When an on-premises system makes sense

A model on the firm's own premises is not the cheapest choice. For a firm of ten people, a business subscription to a cloud assistant costs around 20 euros per person per month; a capable machine, its installation and support cost more, even over three years. It makes sense when what leaves the firm matters more:

• Documents that must not leave. Tax returns, expert reports, extraordinary transactions, anti-money-laundering files: material for which even a well-written contract with a foreign provider leaves a doubt, for the client before the professional.
• Daily use. Searching the firm's archives, summarising deeds and contracts, extracting data from invoices and bank statements, drafting letters: repetitive work on the firm's own documents, where a mid-sized open model is enough.
• Someone to look after it. A model on the premises has to be updated, protected and monitored. The most widespread tools for running it are built without authentication and must be closed off before being put on a network. If there is no one in the firm who does this, a provider is needed to do it.

The cloud remains better suited to occasional use, to reasoning about complex and novel questions, and to web research, where the largest models are still ahead.

The two routes do not exclude each other: many firms will use the management software's assistant for bookkeeping and an on-premises system for the most sensitive documents. In any case the rules ask for the same three things: know which tools are used and with what guarantees, write it down in a firm policy, and tell the clients.

---

Main sources: Fondazione Nazionale dei Commercialisti, Rapporto 2026 sull'Albo; CNDCEC, code of ethics, Art. 21, paragraphs 8–10 (Informativa 178/2025); CNDCEC and FNC, "L'aiuto intelligente al Commercialista", October 2025; ODCEC Torino, "L'intelligenza artificiale nello studio", June 2026; Osservatori del Politecnico di Milano, Professionisti e innovazione digitale, July 2026; Law of 23 September 2025, no. 132, Art. 13.