BStBK adds AI-labelling Q&A for public-facing AI texts

The Bundessteuerberaterkammer has reworked its AI guidance for practices.Bstbk The FAQ “KI im steuerberatenden Berufsstand” now carries the date 27 October 2026Bstbk with a September 2026 legal statusBstbk (previously January 2026 / July 2025), and beneath the typographical overhaul it makes two substantive moves: a new labelling Q&A and a reversed answer on AI-Act duties for firms.

The new question “Müssen KI-generierte Inhalte gekennzeichnet werden?” lays out the Art. 50 Abs. 4 KI-VO labelling dutiesBstbk. Texts generated or manipulated by AI must as a rule be labelled where they are published to inform the public about matters of public interestBstbk — which covers items on the Kanzleiwebsite or on social media. The duty regularly falls away where the content passed human review or editorial control and the firm takes editorial responsibility for publicationBstbk, and purely orthographic or minor stylistic polishing is no generation or manipulation at allBstbk. For image, audio and video content, labelling is required above all for deepfakesBstbk: realistic depictions that can falsely appear to show real persons, events or statements — while ordinary AI illustrations or minor retouching need no labelBstbk. Purely internal use such as working drafts, meeting notes or internal checklists carries no labelling dutyBstbk, but the FAQ warns that any later external publication or unintended sharing demands a fresh legal check. Its tip: set firm rules for review and release, and consider voluntary internal labelling to keep AI use traceable.

The general AI-Act answer no longer tells firms they face no specific duties.Bstbk Steuerberatungskanzleien must now observe certain requirements when deploying AI systemsBstbk: KI-Kompetenz under Art. 4 KI-VOBstbk — firms using AI on their own responsibility are as a rule Betreiber within Art. 3 Nr. 4 — and transparency duties where their conditions are met. Special transparency requirements such as technical labelling attach primarily to providers of generative systemsBstbk, but firms using those systems can themselves face labelling duties depending on content type and useBstbk. Transparency rules including labelling and deepfake duties have applied since 2 February 2025Bstbk, with the comprehensive duties — particularly for high-risk systems — expected from August 2026Bstbk. The FAQ recommends an internal AI policy fixing which tools may be used how, with risk assessment, documentation, monitoring, quality assurance and human oversight.

Client-facing automation keeps its answers from the previous version. Firms offering AI chatbots or auto-responders must as a rule ensure users are told they are interacting with an AI systemBstbk, under the Art. 50 Abs. 1 KI-VO transparency dutiesBstbk. The retained section on joint firm-client AI use — a client portal with AI chat, AI-assisted document checking, shared document platforms — calls for written role allocation in the engagement letter or a supplementary agreement, client transparency on when and why AI is used, data-protection and secrecy safeguards including an Auftragsverarbeitungsvertrag, encryption and access controls, usage rules for clients, and liability terms making clear that final tax assessments always come from the firm.

Unchanged, the data-protection impact paragraph reminds firms: supervisory authorities often require a Datenschutz-Folgenabschätzung under Art. 35 DSGVO where AI use is likely to pose high risks to individuals' rightsBstbk — frequently the case with extensive client data or special categories such as health data — so firms should run a risk analysis before introducing an AI tool and demand functioning and training-data information from the provider when choosing it.

Legal basis: BStBK FAQ “KI im steuerberatenden Berufsstand” (Stand 27 October 2026, Rechtsstand September 2026)Bstbk, whose updated answers turn on Art. 50 KI-VO.