May tax advisers use AI, and what is it worth using for?

What do the professional rules require when a practice uses AI?

The professional rules do not prohibit AI. They govern who may see client secrets and who is accountable for the result. Under § 57(1) StBerG, tax advisers practise their profession independently, on their own responsibility, conscientiously and confidentially; confidentiality covers everything that has become known to them in the course of their professional work. Anyone who discloses another person's secret without authorisation commits an offence under § 203(1) no. 3 StGB (German Criminal Code), punishable by imprisonment of up to one year or a fine. Unlike the GDPR, professional secrecy also protects information about companies, not just about natural persons.

An AI provider is a service provider within the meaning of § 62a StBerG. The provision allows you to give it access to client secrets insofar as this is necessary for the service, and ties this to four conditions:

• Careful selection (para. 2). If compliance with the requirements is no longer ensured, the cooperation must end without delay.
• Contract in text form (para. 3): a duty of confidentiality with instruction on the consequences under criminal law, knowledge only as far as necessary to perform the contract, and a rule for further persons.
• Services provided abroad (para. 4): only if the protection of secrets there is comparable to that in Germany.
• Services for a single client matter (para. 5): only with the client's consent.

Whether an AI tool used for a specific case falls under para. 5 or counts as a general work tool of the practice has not been clearly settled, in the view of the Federal Chamber of Tax Advisers. It advises obtaining consent before entering sensitive client data, and otherwise anonymising or pseudonymising data and documents. Why even a provider's possibility of access counts is explained in the article AI and § 203 StGB.

The three provisions are structured in the same way; the rules on services abroad and single client matters are identical apart from the professional title. How lawyers answer the open questions is covered in the article AI for lawyers.

What do the Federal Chamber of Tax Advisers and the DStV advise?

Both consider AI permissible and useful, but call for clear rules on which tool is used with which data for which purpose. The Federal Chamber of Tax Advisers (BStBK) has published an FAQ catalogue on AI in the tax advisory profession, as of 27 January 2026, reflecting the law as of July 2025. The key points:

• Responsibility stays with the professional. AI results must be checked like the work of an inexperienced employee; the Chamber expressly warns about incorrect citations of Federal Fiscal Court (BFH) decisions.
• Only approved accounts, no private AI accounts for work purposes.
• Public chat services without the client's consent only without information that allows conclusions about the client or the matter.
• Providers from the EU or with data centres in the EU are preferable.
• Documentation according to risk: a lean AI register with tool, purpose, type of data and responsible person is often enough.

In April 2026, the German Association of Tax Advisers (DStV) published a model AI use policy. Its basic logic is “tool – use case – data – risk”: AI may only be used if the tool has been approved, the use case has been described and the requirements of its risk level are met.

Tasks at levels 1 and 2 can be tried out right away. Level 3 requires the contracts under § 62a StBerG and the GDPR.

What AI does DATEV offer today?

DATEV (Germany's standard accounting software) offers AI in several programs. As of September 2026, DATEV lists, among others, these components on its AI overview page:

If you already work with DATEV, check these components first, because the data is already in the system. The BStBK considers AI from practice software vendors preferable, because in its view questions of processing on behalf have been settled there. Still, check for each component which data it processes. Another tool is only worthwhile for tasks your practice software does not cover.

Which tasks in a tax practice is AI worth using for?

AI brings the most benefit where a draft or a list is produced that a professional then checks. The BStBK names as typical areas of use: receipt recognition, data extraction, drafts of submissions, emails and client information, summaries of case law and administrative circulars, and checklists. Classified by suitability and data issue:

A special case is credit card statements or statements from foreign banks that only come as PDFs: an AI agent can turn them into a table that you check and import. How companies themselves use AI in accounting is covered in AI in accounting.

Where does AI not help in tax advice?

AI replaces neither the tax assessment nor the final check. The typical errors can be caught with a fixed checking routine:

• Invented references. Language models cite judgments, case numbers and margin numbers that do not exist.
• Outdated law. DATEV states that the Copilot without additional sources has knowledge up to the end of September 2024; later changes in the law are then missing.
• Calculation errors. Spot-check totals and tax calculations.
• Scans without a text layer need text recognition (OCR), which not every AI tool includes.
• Never take deadlines from an AI text; the practice's deadline control remains authoritative.

How does a tax practice start a pilot without client data?

Start with two or three tasks without client data, one responsible person and written rules. The BStBK also recommends starting with clearly defined pilot cases.

1. Choose tasks

   Take tasks at DStV levels 1 and 2, such as a newsletter, a document checklist or a work instruction. Measure beforehand how long they take today.

2. Check and approve the tool

   Go through the list below, even for free tools, and use practice accounts.

3. Set up rules and an AI register

   Which data may go in, who checks, where is it documented? The DStV model policy and the model register in the BStBK FAQ catalogue are suitable templates.

4. Train the team

   Art. 4 of the EU AI Act requires deployers to take measures that promote their staff's AI literacy. What training can look like is covered in the article AI literacy training.

5. Decide after four weeks

   Compare time spent and errors with the situation before. Only after that do tasks with client data follow, and only with a contract under § 62a(3) StBerG.

What does the EU AI Act require of a tax practice?

Little, as long as the practice only uses AI tools. It is then a deployer within the meaning of the Act, and in the BStBK's view, ordinary use for research, bookkeeping support or texts does not fall into the high-risk categories. The most important obligation is in Art. 4, which Regulation (EU) 2026/1744 reworded with effect from 27 July 2026: deployers take measures to support the development of their staff's AI literacy, and do not have to guarantee a particular level. For chatbots that clients write to directly, the transparency obligations under Art. 50 have applied since 2 August 2026.

Practice templates yes, client data only after checking: Biwak in a tax practice

Today, Biwak suits practice tasks without client data or with anonymised files; it has no connection to DATEV. The desktop app works in a folder that you assign to it, such as the folder with your practice templates: Word templates, Excel and CSV exports, PDFs with a text layer. You describe in one sentence what should be produced from them, and the AI agent saves the result as a new file alongside, reliably as CSV, HTML or Markdown. If revising a model template goes wrong, you can restore the previous state with one click, because Biwak backs up the folder before each task.

Biwak works with DATEV via files: you put an export, such as a CSV file, into the working folder, and Biwak creates a list or analysis from it; nothing goes back to DATEV automatically. For the assessment under § 62a StBerG, the actual data route is what counts, because a local app on its own does not meet any requirements of professional secrecy. With every model task, your question, the necessary conversation context and the file contents used go via Biwak to the model provider. Processing happens in the EU: at Microsoft Azure in the EU Data Zone, or as a fallback via Prem AI (contracting party PREM SA, Switzerland). For careful selection under para. 2 and protection abroad under para. 4, these circumstances currently count: at Azure, Microsoft may store suspicious requests for abuse detection and have them reviewed by humans, and according to Microsoft the EU Data Zone can also include EFTA states such as Norway and Switzerland. If Prem steps in because Azure rejects a task, the following applies: Prem names its compute partners Nebius and TensorX, but a DPA with Prem has not yet been signed; its commitment not to store inputs rests, by its own account, on contract and operating procedures rather than technical safeguards; and Biwak does not currently use the encrypted mode that Prem recommends for legal documents. Biwak's DPA contains a confidentiality undertaking in text form with a notice of criminal liability. Even so, client data does not belong in it without your own assessment under § 62a StBerG and the GDPR; the details are in the privacy policy.

Frequently asked questions