What sources does Taxxa search?

Taxxa searches a continuously updated collection of more than one million pages from tax authorities, government publications, legislation and case law across the European jurisdictions it covers. Every response cites the underlying material, so a professional can inspect the rule, its jurisdiction and whether it is current before relying on the result.

Can Taxxa work with a team’s own documents?

Yes. Research can combine the public knowledge base with a team’s own PDF, DOCX, CSV and TXT files in one workspace. That makes it possible to compare a client document with the relevant guidance, trace a conclusion to its source and prepare a draft without losing the evidence that supports it.

Who is Taxxa built for?

Accountants, auditors, tax advisors, payroll specialists, bookkeepers and advisory teams working across European jurisdictions. Shared workspaces keep research available to colleagues, and citations make review easier.

Where is customer data stored, and is it used to train models?

Customer data stays in the European Union and is never used to train language models.

Does Taxxa replace professional judgement?

No. Taxxa is a research and drafting workspace. It helps professionals reach an informed starting point faster; the professional still reviews the evidence and makes the final decision. It does not file returns, post bookkeeping entries or give legal advice.

SwedenFAR

High AI Act fines do not automatically mean a material breach

FAR's rewritten AI explainer says high AI Act fines do not automatically make a breach material — assess facts and effects under ISA 250 — and sets staged AI Act dates through 2028.

By Taxxa AI OyPublished 27 October 2026

IT, Cybersecurity & Data Audit

FAR has rewritten its explainer on artificial intelligence in accounting, payroll and audit.FAR The text of 27 October 2026 treats AI as a support tool whose output must be checked against agreements, source documents and verifiable calculations. Professional scepticism and the auditor's responsibility for sufficient appropriate evidence remain unchanged.

In accounting, AI may interpret invoices and propose postings, but a proposal for a service spread over time must still be reconciled with the contract, the correct period, value-added tax and the chart of accounts. Authorisation and other controls stay in place, and a recurring but wrong proposal must not be accepted only because similar items were processed. In payroll, a tool may structure recurring questions or draft an explanation from anonymised material, but the agreement version, dates and conditions must be checked first. Removing a name is not always anonymisation.

In audit, deviation analysis may direct further procedures, but an unusual posting is not automatically an error and a normal pattern does not prove correctness. Analysis of recorded transactions does not show that the population is complete. Completeness and accuracy of the records must still be checked and the need for additional evidence assessed.

A well-formulated answer is not reliable evidence. Prompting improves the starting position but does not transfer responsibility. Hallucinations and bias can be systematic, and confident language does not measure reliability. Material amounts must be recalculated by a controllable method with inputs, method, rounding and interpretation documented.

Client information must go only to approved services under internal policy. Roles under the General Data Protection Regulation and the legal basis must be mapped, data minimised, processor agreements concluded where needed, and access, sub-processors, storage, erasure and any transfers outside the EEA reviewed, with a risk assessment and, where required, an impact assessment. That data is not used for training does not by itself make a service safe. Agents with connected tools receive only necessary permissions, with human approval before steps such as payments or register changes. Untrusted instructions in documents read by the system are treated as data, not as authorised instructions.

On Regulation (EU) 2024/1689, the AI Act, the explainer gives 2 August 2026 as the general date of applicationEuropa, with Chapters I and II applying from 2 February 2025 and Chapter V on general-purpose AI models from 2 August 2025. The central high-risk duties in Chapter III, Sections 1 to 3, apply from 2 December 2027 for systems under Article 6(2) and Annex IIIEuropa and from 2 August 2028 for product-related systems under Article 6(1) and Annex IFAR. Ordinary accounting or audit use is not automatically high-risk. Article 4 requires measures supporting AI literacy of staff and others using the systems, having regard to knowledge, experience, task and context, not a guarantee of a given individual level. Transparency duties for certain systems sit in Article 50, model duties in Article 53 and additional duties for models with systemic risk in Article 55. The AI Act does not replace the General Data Protection Regulation.

On suspected breaches of law, high potential sanctions may be relevant but do not automatically make a matter materialFAR. Facts and possible effects on the financial statements must be assessed under ISA 250, paragraphs 19 to 22 and A19.FAR NOCLAR belongs to the IESBA Code of Ethics and complements ISA 250, and the ethical framework is not limited to situations with material misstatements.

Legal basis is Regulation (EU) 2024/1689 as amended by Regulation (EU) 2026/1744Europa, ISA 250 and the IESBA Code of Ethics (NOCLAR).

Check AI-supported bookkeeping, payroll and audit steps against source documents and verifiable calculations, and assess any suspected AI Act breach by its facts and financial-statement effect under ISA 250 rather than by the fine level alone.

Sources