AI for tax advisers & tax firms: § 203 StGB, receipts, clients | meinGPT
How tax firms use AI while preserving their duty of confidentiality (§ 203 StGB): pre-capture receipts, draft client communication, research tax law with sources — with real workflows, an example prompt, strict review limits and selection criteria for professionals bound by secrecy.
For Tax advisers, firm owners, tax clerks and firm management.
- Who it is for
- Tax advisers, firm owners, tax clerks and firm management
- Impact
- Less time in pre-capture and routine correspondence — every tax-relevant statement stays reviewed and owned by the tax adviser
- Task
- Pre-capture receipts, draft client communication and research tax law — while preserving the duty of confidentiality (§ 203 StGB)
AI in tax advice means using generative AI for the preparatory, recurring tasks of a firm — extracting data from receipts and invoices, drafting client correspondence, researching tax law with sources and making firm knowledge accessible. The decisive frame is the professional duty of confidentiality: tax advisers in Germany are professionals bound by secrecy under § 203 StGB and may only pass client data to external IT or AI providers if those providers are involved as a 'contributing person' within the meaning of § 203 (3) sentence 2 StGB and the service contract meets the professional requirements of § 62a StBerG (text form, careful selection, a confidentiality obligation — plus, for services that directly serve an individual mandate, the client's consent under § 62a (5) StBerG) — on top of EU processing, a DPA and the assurance that input is not used for training. meinGPT offers tax firms exactly this separate confidentiality undertaking under § 203 StGB: in it, SelectCode commits in writing to secrecy as a contributing person under § 203 (3) sentence 2 StGB. The AI supports the preparation; every tax-relevant statement is reviewed and owned by the tax adviser.
From the task to productive AI use
A firm assistant extracts structured fields from uploaded receipts and invoices — supplier, date, amounts, tax rates, line items — and proposes a pre-capture that a clerk checks against the original document and approves. For client communication it drafts factual text from bullet points (deadline reminders, requests for missing documents, explanations of assessments), which the tax adviser reviews before it goes out. For tax-law research it summarises sources with citations — as preparation, not as binding tax advice from the AI. Firm knowledge (checklists, internal instructions, client handbooks) becomes queryable through a knowledge base. Because several leading models are available behind one interface, you can pick the right one per task; permitted data sources can be connected via connector/API. The prerequisite for all of it is an operation that preserves the duty of confidentiality — EU processing, a DPA, no training on input and the provider's written confidentiality undertaking.
- Who it is for
- Tax advisers, firm owners, tax clerks and firm management
- Impact
- Less time in pre-capture and routine correspondence — every tax-relevant statement stays reviewed and owned by the tax adviser
- Task
- Pre-capture receipts, draft client communication and research tax law — while preserving the duty of confidentiality (§ 203 StGB)
What Tax advice / tax firm gets done with AI
Concrete, repeatable flows — from the first prompt to a dependable result.
Pre-capture receipts and invoices
From uploaded documents the assistant extracts structured fields — supplier, date, net/gross, tax rate, line items — and proposes a pre-capture. Tax clerks check every value against the original document and approve; the AI takes over the typing, not the professional responsibility. The documents stay inside an environment in which the duty of confidentiality is preserved. Because this pre-capture directly serves an individual mandate, the client's consent under § 62a (5) StBerG is part of the precondition.
Draft client communication
From bullet points, the assistant drafts factual text for recurring correspondence — deadline reminders, requests for missing documents, explanations of assessments. The tax adviser reviews content and tone before it goes out; the AI supplies the draft, not the binding statement to the client.
Research tax law with sources
As preparation, the assistant summarises sources and provisions with citations, so the professional review can start faster. This is research support, not binding tax advice from the AI — the legal and tax assessment is always made and owned by the tax adviser.
Prepare deadlines and advance returns
The assistant helps structure recurring work around deadlines and advance returns — producing checklists and reminder text from internal guidance, for instance. The calculation, review and filing themselves stay the responsibility of the firm and its specialist systems, not the AI.
Make firm knowledge accessible
An assistant connected to internal checklists, working instructions and client handbooks answers the team's recurring questions with a source reference — new staff find the right approach faster, without knowledge being locked in individual heads. Access follows the permissions granted.
Build a firm assistant and bake in the review steps
Firm management assembles an assistant with its own templates, tone of voice and a fixed review reminder ('double-check every tax-relevant statement') without code and releases it to the team — so human control is part of the workflow, not optional.
A real prompt, a real answer
Nothing hidden — you see the input and the result before you sign up.
Draft a polite, factual email to a client who is still missing documents for the annual accounts: Q4 bank statements, two incoming invoices of €5,000 each, and the hospitality receipt of 12/11. Ask for them by the end of next week, friendly tone, no tax assessment. Flag the points I should check before sending.
| Element | AI draft | Status |
|---|---|---|
| Subject | Missing documents for your 2025 annual accounts | Draft — please check |
| Salutation | Dear Ms …, | Draft — please check |
| Core request | "…to complete them we are still missing: Q4 bank statements, two incoming invoices of €5,000 each, hospitality receipt of 12/11…" | Draft — please check |
| Deadline | "…by the end of next week…" | ⚠ insert and check the concrete date before sending |
| Tax assessment | — (deliberately none) | ✓ contains no binding statement |
| Sign-off | Kind regards, … | Draft — please check |
Put it to work in your own company
In a short live demo we show how this solution runs in your company with meinGPT, GDPR-compliant — using your own use cases.
Or download the AI for professionals bound by secrecy — the § 203 selection criteria (PDF, German) as a PDF:
A work email is enough — processed in line with the GDPR.
Built for enterprise compliance
Tax firms process professional secrets, and the operation is designed accordingly. meinGPT is operated by SelectCode GmbH, which is ISO 27001 certified and has its security reviewed regularly through independent penetration tests (most recently SySS, 2025). Processing takes place in the EU, a data processing agreement (DPA) is standard, and input is not used to train the models. Access to documents, client data and the knowledge base follows strict permissions through central permission management with SSO, is limited by least-privilege scopes and is traceable through audit logs — important for audit readiness and evidence. For professionals bound by secrecy the decisive building block is added: meinGPT offers tax firms a separate confidentiality undertaking under § 203 StGB. In it, SelectCode commits contractually and in writing to secrecy as a contributing person under § 203 (3) sentence 2 StGB — going beyond the text form that § 62a (3) StBerG requires for the service contract — keeping entrusted secrets confidential (including beyond the end of the contract), obtaining only the knowledge necessary to perform the contract, and binding all deployed staff and any sub-contractors to confidentiality in writing as well (undertakings to be produced on request). Together with EU processing, a DPA and the exclusion of training on input, that is the basis for using AI while preserving professional confidentiality; meinGPT provides the undertaking on request. Client data therefore stays under control in the firm's environment rather than being processed through private AI accounts.
- § 203 StGB / § 62a StBerG: Does the provider commit contractually to confidentiality as a contributing person — with a separate confidentiality undertaking — and does the service contract meet the text form required by § 62a (3) StBerG?
- § 62a (5) StBerG: Is it settled for which tasks the client's consent is obtained — required as soon as the service directly serves an individual mandate?
- EU processing & DPA: Is client data processed within the EU and is there a data processing agreement (Art. 28 GDPR)?
- No training on input: Is it contractually assured that prompts and uploaded receipts are not used to train the models?
- Client separation & permissions: Can access be limited per client and role (least privilege), so only authorised people see client data?
- Evidenceability: Is there central permission management, SSO and audit logs to make access traceable for review and audit?
- Human control: Does the setup support a four-eyes principle in which every posting- and tax-relevant statement is reviewed by the tax adviser?
- Adoption: Is there training and are there champions, so the firm's team uses AI correctly, with review and in line with data protection?
What this solution cannot (yet) do
Honesty is part of the solution. These limits are known — and therefore plannable.
Confidentiality first: client data may only go into an AI environment in which the provider is bound to confidentiality as a contributing person under § 203 (3) StGB — never through private AI accounts (shadow AI), as that can amount to a criminally sanctioned disclosure.
Every posting- and tax-relevant statement must be reviewed and owned by the tax adviser — the AI may pre-capture, draft and research, but must not post, calculate deadlines or advise bindingly on its own.
AI models can misread or invent amounts, tax rates, sources or fields; extracted values and cited provisions must always be checked against the original document or the primary source, and a four-eyes principle is strongly recommended.
The AI does not replace tax advice: binding tax and legal assessment, account assignment and compliance with GoBD and tax law remain the firm's professional responsibility.
Frequently asked questions
Yes — but only while preserving the duty of confidentiality. Tax advisers are professionals bound by secrecy under § 203 StGB. External IT and AI providers count as 'contributing persons'; client data may be made accessible to them under § 203 (3) sentence 2 StGB insofar as this is necessary for the professional activity; professional law then requires, under § 62a StBerG, a service contract in text form with careful selection and a confidentiality obligation — and, for services that directly serve an individual mandate, the client's consent under § 62a (5) StBerG. On top of that you need EU processing, a DPA and the assurance that input is not used for training. meinGPT provides tax firms with a separate confidentiality undertaking under § 203 StGB, in which SelectCode commits in writing to secrecy as a contributing person under § 203 (3) sentence 2 StGB. Client data must not be processed through private AI accounts without such a binding commitment.