AI for accountants and labor consultants: uses, limits and rules

01 What AI can do in a firm, on the firm’s own material

Think of AI not as “a tool that knows tax law” but as an assistant that has read your documents: circulars, internal procedures, the deadline calendar, client files, the answers you have been giving to the same questions for years. The more the answers start from there, the easier they are to check.

• Searching the firm’s circulars. “How do we handle mileage refunds for partners?”: the answer comes from the circular you already wrote, with the reference to open it.
• Answering recurring client questions. Documents for the tax return, this month’s deadlines, what is needed to hire an apprentice: AI prepares the first answer from your material, a colleague checks it.
• Summarizing a file before a meeting. What was decided, which documents are missing, which deadlines are open, including those in the firm’s deadline calendar.
• Drafting in the firm’s style. Letters to clients, notices to a client’s employees, minutes: the draft starts from your templates, the signature stays yours.
• For labor consultants, collective agreements and payroll procedures. The rules of the agreement applied to a client, the steps of a hiring the way your firm does it.

This is what firms that have adopted AI already do. According to the Politecnico di Milano observatory on professional firms, the most common use is searching legislation and case law, the second is summarizing documents; chatbots answering clients are still rare.

Share of the firms that use AI for each task. Source: Osservatorio Professionisti e Innovazione Digitale, Politecnico di Milano, press release of 2 July 2026.

02 How many firms already use it?

It depends on what you measure. The first figure is about firms: in 2026 the Politecnico di Milano found that AI tools chosen by the firm (features already built into practice management software don’t count) are used by 39% of accounting firms and 42% of labor consulting firms. Across all professions, the share goes from 27% of the smallest firms to 62% of the large ones.

The second is about people. The survey of the accountants’ national research foundation (more than 4,000 responses, collected between July and September 2025) found that 7.5% of accountants use AI a lot, 27.6% quite a lot, 47% a little and 17.9% not at all; within three years, those expecting to use it a lot or quite a lot rise to 71.9%. The risks they cite most: uncertain regulation (31.9%) and liability for errors generated by AI (26.6%).

Together they say that almost everyone has tried it (more than eight accountants in ten, counting those who use it a little), but only a minority of firms has turned it into a way of working, with chosen tools and rules.

03 What stays with the professional

Law 132/2025 says it in one sentence: in the intellectual professions, AI is there “solely for ancillary and support activities”, and the professional’s intellectual work must prevail (art. 13, paragraph 1). This is not a new idea: the Italian civil code requires professionals to carry out the engagement personally and lets them rely on assistants “under their own direction and responsibility” (art. 2232). AI is help, not a substitute.

The accountants’ code of ethics, amended by their national council (CNDCEC) on 20 November 2025, is more specific. It forbids using AI “in such a way that it replaces their own intellectual activity” or the assessment and interpretation of the facts and rules (art. 21, paragraph 8). And it requires checking sources and data, using tools that are secure and comply with data protection law, and making sure the whole firm, employees and collaborators included, understands the technology it uses (paragraph 9). Using AI never excuses anyone from the duties of the code and of the law.

For labor consultants, their national council (the CNO) wrote to the provincial councils on 6 October 2025: AI is to be used only as support, and responsibility stays personal, neither reduced nor transferred because a tool was used.

In practice, these stay with the firm:

• interpretation: which rule, which collective agreement, which tax treatment;
• assessing the client’s facts, which AI only knows from the documents you give it;
• the decision and the signature, on an opinion, a return, a payroll calculation;
• the relationship with the client, including telling them how you work.

04 Which rules apply: Law 132/2025, codes of ethics, GDPR

There are several layers, and they add up.

The law sets no form for the client notice: it says the information serves “to ensure the relationship of trust” and must be clear, simple and comprehensive. On 9 October 2025 the accountants’ national council published a model clause for engagement letters, which other firms can adapt too. In four points: AI only for ancillary purposes (research, drafts, content that decides nothing); no automated decisions about the client; on request, information on the tools and security measures; and the client can ask, with a reasoned request, that AI not be used on their engagement.

For accountants, breaking these rules has a written price. The sanctions code, updated at the same council meeting, provides for suspension of up to six months for breaching paragraphs 8 and 9 and up to three months for not informing the client (paragraph 10), from 21 November 2025.

Then there is privacy. Firms process their clients’ data and, for labor consultants, their clients’ employees’ data, which can include health information (sick leave) and trade union membership (union dues deducted from pay): special categories under the GDPR (art. 9). If an AI provider processes that data on the firm’s behalf, a contract under article 28 (the DPA) is required. If the firm itself processes the data as a processor for the client, for example because the client appointed it as such for payroll, the AI provider becomes a sub-processor and the client’s prior written authorization is required (art. 28(2)).

The accountants’ council guide of October 2025 draws the line that matters most. Free or low-cost versions of chatbots typically allow uploaded content to be used to train their models, and uploading a client’s confidential document to such a service, the guide says, “would amount to a breach of professional secrecy and of the GDPR”. Business versions rest on contractual commitments instead: the data belongs to the customer and does not train the models.

Labor consultants also face the other side: the same law (art. 11) requires employers to inform workers when they use AI, in the cases and ways set by article 1-bis of Legislative Decree 152/1997. Clients will bring the question to the firm: it helps to have a standard answer ready.

Finally, the AI Act: since 2 February 2025 a firm that uses AI systems (a “deployer”, in the Regulation’s terms) must take measures to ensure a sufficient level of AI literacy among its staff (art. 4). We wrote about it in You are already an AI deployer.

05 How to start: a plan for the first month

The starting point is not buying a tool: it is deciding what AI may read, who uses it and how its output is checked. According to the Politecnico di Milano, among firms using AI about half have a procedure to check its output before it reaches the client, fewer than one in four regulate access to the tools by seniority, and few pay attention to where the data is stored. The first month is for closing those three gaps.

1. Week 1: map what is used. Ask, with no consequences for anyone, who uses which tool and for what. You will find personal accounts, free versions, client documents uploaded. The point is to know, not to punish.
2. Week 1: choose tools that come with a contract. A provider that signs a DPA, doesn’t use your content to train its models and states where it stores the data. Personal accounts stop being used for client work.
3. Week 2: update the engagement letter. The AI notice in new engagements and a short message to current clients. The accountants’ model clause is a good starting point.
4. Week 2: start from the firm’s material. Circulars, procedures, templates and the deadline calendar hold your know-how and little personal data. Client files come later, with permissions and contracts in place.
5. Week 3: one use case at a time. For example the twenty questions clients ask most: check that the answers come out of your own circulars, with the reference to the document.
6. Week 3: write down the review rule. Nothing reaches a client without a professional reading it; who checks what, and where the record is kept. The accountants’ guide also proposes a register of the AI systems in use.
7. Week 4: training on your own cases, and one page of rules. Twenty real examples, each with a good and a bad answer, teach more than a generic course. Then a one-page policy: what may be uploaded, with which tools, who checks. The same guide includes a template.

This is also how ekory works: Ekos, ekory’s AI agents, read only the documents the firm chooses to give each of them, answer with the document and page, and the archive is stored in Europe without training the models. There are concrete examples on the pages for accounting firms and labor consultants, and answers for your data protection officer on the security page.

06 Note on sources

This article is not legal advice. It reports what the rules and the professional bodies’ documents say as of 10 October 2026; AI rules change quickly, and your own situation needs a professional who knows your firm.

• Laws: the text in force on Normattiva (Law 132/2025, article 2232 of the civil code) and EUR-Lex (GDPR, AI Act). Translations into English are ours.
• Accountants’ code of ethics and sanctions code: the texts published by the national council on 4 December 2025 (Informativa 178/2025).
• Data on firms: Politecnico di Milano press release of 2 July 2026; the shares by use refer to firms that use AI.
• Data on accountants: preview of the survey by the Fondazione Nazionale di Ricerca dei Commercialisti (22 October 2025). The release gives 34.1% for “a lot or quite a lot”, but the two items it reports (7.5% and 27.6%) add up to 35.1%: we use the items.
• Labor consultants: we found no survey by their professional body that we could verify at the source; we use the Politecnico di Milano data, broken down by profession.

Frequently asked questions

Sources

1. Legge 23 settembre 2025, n. 132, art. 13 (intellectual professions), Normattiva, 2025.
2. Legge 23 settembre 2025, n. 132, art. 11 (AI in employment), Normattiva, 2025.
3. Codice deontologico della professione, amended on 20 November 2025, CNDCEC, 2025.
4. Codice delle sanzioni disciplinari, updated on 20 November 2025, CNDCEC, 2025.
5. Informativa n. 178/2025, amendments to the code of ethics and the sanctions code, CNDCEC, 4 December 2025.
6. Model contract clause for informing clients about AI, CNDCEC press release, 9 October 2025.
7. Guida operativa di intelligenza artificiale #3, L’aiuto intelligente al commercialista, CNDCEC and Fondazione Nazionale di Ricerca dei Commercialisti, 24 October 2025.
8. Utilizzo dell’intelligenza artificiale nelle professioni intellettuali, note of the Consiglio Nazionale dell’Ordine dei Consulenti del Lavoro, 6 October 2025.
9. Commercialisti e IA: preview of the survey “Organizzazione dello studio e impatto dell’intelligenza artificiale”, Fondazione Nazionale di Ricerca dei Commercialisti, PRESS, 22 October 2025.
10. La spesa ICT degli studi professionali in Italia supera i 2 miliardi, Osservatorio Professionisti e Innovazione Digitale, Politecnico di Milano, 2 July 2026.
11. Regulation (EU) 2016/679 (GDPR), articles 9 and 28, EUR-Lex, 2016.
12. Regulation (EU) 2024/1689 (AI Act), articles 4 and 113, EUR-Lex, 2024.
13. Codice civile, art. 2232, Normattiva.