Legal AI in Austria: Providers Compared 2026

The Austrian market for legal AI holds around a dozen serious providers in 2026, from Vienna start-ups through the publishing houses MANZ and LexisNexis to international platforms with no Austrian footprint at all. Anyone selecting one is really deciding three things: which legal sources are covered, whether the professional duties under section 40 RL-BA 2015 are met, and whether the tool merely researches or also reviews.

This overview sorts the market along those three questions. It was compiled in August 2026 from every provider with a demonstrable Austrian connection, together with the search results Austrian law firms actually encounter.

Why this overview did not exist before

Anyone searching in Austria in August 2026 for a comparison of legal AI tools will not find one. Results for "Legal AI Österreich" or "KI für Anwälte Österreich" consist of provider home pages, conference announcements, seminar listings and a university thesis. For the query "Legal AI Vergleich Österreich", fourth place goes to a German overview covering sixteen tools against the BRAK professional rules: no RIS, no OGH, not a word on the Austrian Rechtsanwaltsordnung.

That is the reason for this text. German overviews are not wrong for the Austrian market, but they answer the wrong questions. A tool fluent in German case law does not thereby know OGH judgments, and a provider that satisfies the BRAO has said nothing about section 9 RAO.

What actually decides the choice in Austria

Selection criteria for legal AI in Austria follow from professional rules, not from product brochures. Section 40 para. 3 RL-BA 2015 expressly governs when a lawyer may engage external service providers for electronic data processing. Five conditions must be met together:

1. The client's interests are protected.

2. The lawyer selects the external provider carefully.

3. The provider is contractually and verifiably obliged to inform the lawyer without delay in the event of a search of the premises.

4. Technical and organisational measures reflecting the state of the art are in place.

5. The client is informed about the categories of providers engaged and the services they render.

These five points are the most useful assessment framework the Austrian market offers, and they appear in no provider comparison. Two of them can be checked very concretely before any purchase. Point 3 requires a contractual commitment to report a search of the premises, which almost nobody asks about. Point 5 presupposes that the provider names its sub-processors at all, because without that list a firm cannot inform its clients.

What such a commitment looks like can be shown from our own terms. The Austria annex to the CASUS terms and conditions carries it verbatim: the company undertakes to inform the customer without delay in the event of official seizure, a search of the premises or any other official access, as far as legally permitted, with express reference to section 40 para. 3 no. 3 RL-BA 2015. What matters is not that this one provider gives it, but that the commitment sits in the contract rather than in a brochure. Anyone comparing should ask every provider to show it.

The background is the duty of confidentiality itself. Under section 9 para. 2 RAO a lawyer is bound to secrecy regarding matters entrusted to them, and that duty expressly extends to auxiliary staff. Section 40 para. 2 RL-BA 2015 requires a "verifiable contractual transfer" of the confidentiality obligation, and does so "including every form of electronic data processing". A legal AI provider is therefore not ordinary software but a third party handling client matters.

This is flanked by the anti-circumvention rule: section 9 para. 3 RAO and section 157 para. 2 StPO prohibit circumventing the right of confidentiality through official measures, expressly including the seizure of data held on storage media. Only against that background does the reporting duty in point 3 stop looking like a formality.

What does not decide the matter is the number of features on a product page. Almost every provider in the Austrian market can research, summarise and draft. The differences lie elsewhere.

Data protection comes before functionality

In first conversations with law firms, questions almost never start with features but with data protection and security. One question recurs so reliably that it stands for the whole category: whether documents must be anonymised before upload. The answer depends entirely on the provider. Where zero data retention is contractually agreed and processing takes place in Switzerland or the EU, prior anonymisation is unnecessary. Where content flows into a model whose operator caches it or uses it for improvement, anonymisation changes little.

Source coverage: who holds which law

The question of sources divides the Austrian market more sharply than any feature list, and it is rarely answered honestly. "AI for Austrian law" can mean that the entire federal legal information system including supreme court case law is searchable, or that a language model draws on the content of a single publisher.

RIS (Rechtsinformationssystem des Bundes) is the official primary source for Austrian law and is freely accessible. Advertising the use of RIS sources therefore names the raw material, not an advantage. What matters is what lies beyond it and how deep: the case law of the OGH (Supreme Court), the VwGH (Administrative Court) and the VfGH (Constitutional Court), plus specialist sources such as the finance ministry's Findok, decisions of the financial market authority and of the competition authority.

Two examples show the range. MANZ Genjus KI names its basis more precisely than most: around 61 million text passages from its own legal database, the partner publishers Facultas and Kitzler, plus RIS with federal and state statutes and decisions of all courts, Findok and EUR-Lex. Anyone looking for breadth of sources will find it here. At the other end, publisher access is itself a genuine advantage: MANZ holds commentary literature no AI provider can replicate, and LexisNexis offers briefings and practice content of comparable depth.

To put the scale in perspective: queried through the federal government's official interface, RIS held around 895,000 judicial decisions in August 2026, roughly 357,000 of them from the VwGH, 289,000 from the Federal Administrative Court, 138,000 from the ordinary courts and 24,000 from the VfGH. Anyone claiming RIS coverage should be able to say which part of that they mean.

For Austria, CASUS covers the entire RIS plus OGH, VwGH, VfGH, Findok, the financial market authority, the competition authority and statutes. Research draws on statutes and case law rather than general web sources and returns citations that can be reused. What it does not include is commentary literature, and that is no detail: for questions that cannot be answered seriously without a commentary, the publisher's database remains necessary. The realistic arrangement in an Austrian firm is therefore coexistence, not replacement.

The providers at a glance

The table covers every provider with a demonstrable Austrian connection, checked against each provider's own website in August 2026. "On request" means no public list price could be found; "reported" means the figure comes from trade coverage rather than the provider. Few providers publish source counts, so a number appears only where the provider states one itself.

Three observations follow.

Among providers without Austrian roots, exactly one names RIS. BEAMON AI states a RIS integration on its Austrian page, covering statutes, regulations and case law, and has signed the ÖRAK checklist for AI providers. The large international platforms do not. With Libra this can be checked precisely: its sources page lists legal content by country, eleven of them, from Germany through Switzerland to Hungary, and Austria is not among them. Legora names the US, the UK, France, Germany, Scandinavia, Portugal, Turkey, Australia and Singapore, and likewise not Austria. Harvey names no jurisdiction at all. At the same time both already sit inside the document: Harvey offers add-ins for Word and Outlook, Legora a Word add-in with its own playbooks. The tools are where the work happens; Austrian law is not. That stands out all the more because both are already working inside Austrian houses: Harvey and Legora each name Schönherr on their customer pages, and Legora adds STRABAG and, since December 2025, Erste Group.

Price transparency splits the field exactly in two. CASUS, AI:ssociate, RechtGPT, BEAMON AI, RIDA Online and Libra publish a list price. MANZ-Noxtua, MANZ Genjus KI, JuridicAlly, Lexis+ AI, Harvey and Legora point to sales, which is to say the publisher products and the international platforms. Legora moved its top product to consumption-based pricing in June 2026 but likewise names no amount. For a smaller firm that is a selection criterion in itself, since the effort of an enquiry often is not worth it. An amount without volume caps can also be scaled straight to the size of the firm; a volume-based tariff only after a month on live files.

Volume caps belong in the calculation. AI:ssociate's metered model covers 25 research queries, two drafts and five uploads per month on the entry tier. For a firm running three active matters in parallel, that is the calculation made after the first month, and it decides the real cost.

What the ÖRAK offers include, and what follows them

The Austrian Bar Association currently promotes two AI offers at once. AI:ssociate is free for three months under an ÖRAK cooperation, BEAMON AI free for six months since March 2026 with a 20 % standing discount thereafter. Both announcements rank prominently in Austrian search results, the BEAMON item even first for "KI für Anwälte Österreich".

For a firm this is a good entry point and a poor basis for a decision. A free period answers whether a tool is usable. It does not answer what it costs once embedded in daily work, where the data sits, or whether it handles the task that actually hurts.

The most consequential point usually falls by the wayside during such trials: firms that evaluate with sample documents or an old NDA never see the benefit, because it is precisely the real matters the tool is meant to help with. That is not an adoption problem and not a training gap but a flaw in how the trial is set up. Anyone using the free months should fill them with live files.

Research, review and benchmarking are three different jobs

The Austrian market is almost entirely research-centred. That is historically explicable, since the publishers come from the database side and the start-ups compete against them. But it obscures the fact that legal work on a document consists of at least three distinct jobs requiring different tools.

Research answers a legal question from statute and case law. Every provider in this overview does this to some degree, and here source coverage decides.

Risk review takes a specific contract and names what is problematic about it. The difference from research is fundamental: the review is not abstract but conducted from the perspective of one of the parties, with attribution, relevance and severity for each finding, and with wording suggestions that can be adopted. No provider in the Austrian market advertises structured severity review in this form. At CASUS such a review takes one to three minutes and returns around twenty findings on average; details under Risk Review.

Benchmarking against a standard answers a third question: does this document match what the firm or company has set as its standard? A benchmark checks against an internal playbook or against best practice, shows missing topics and incomplete provisions such as liability without a cap or confidentiality without a deletion duty, and reports the match as a percentage. Here too the Austrian offering currently holds no equivalent.

This distinction is the practical core of the decision. Anyone mainly resolving legal questions is well served by a strong research tool plus publisher access. Anyone reviewing, negotiating and benchmarking contracts daily chooses on different criteria and should ask where that work happens. At CASUS around 80 % of usage takes place in the Word add-in rather than the web application, which confirms a plain observation about legal work: it happens inside the document.

Anyone whose requirements are not yet sharp will find the questions worth settling before a demo in our legal AI procurement checklist. The comparison for the Swiss market also shows how differently two neighbouring legal markets develop.

Choosing by starting position

For a sole practitioner or small firm, price is the first criterion and volume caps the second. The bar-subsidised offers are a sensible entry point, provided the follow-on price is known before the decision. Enterprise platforms with minimum seat counts are out.

For a mid-sized commercial firm, the weight shifts to document work. What matters is Word integration, review from a party's perspective, and whether the firm's own templates can be stored as the standard. Publisher access remains alongside.

For an in-house legal team, the professional-rules question looks different, because section 9 RAO does not apply; the GDPR, the Austrian Data Protection Act and internal policies take its place. Throughput on routine requests, consistent contract terms and the analysis of entire document sets, for instance in due diligence, matter more. For this starting position, bulk analysis with self-defined extraction fields is often a stronger lever than single-case research; more under in-house teams.

For all three, the same check applies before signing: run through the five conditions of section 40 para. 3 RL-BA 2015, in particular the duty to report a search of the premises and the list of sub-processors. Anyone looking for a starting point will find the details on hosting, encryption, zero data retention and professional rules on our security page.

Anyone wanting to form their own view can trial CASUS free for fourteen days with no minimum term. That is only worthwhile with real files from live work, though: a trial with a sample document reliably shows that the software functions, and just as reliably fails to show whether it helps. Access can be set up in a few minutes.

FAQ

Which legal AI providers genuinely cover Austrian law?

As of August 2026 only three are fully localised: MANZ-Noxtua, AI:ssociate and JuridicAlly. Alongside them are providers with Austrian source coverage but no dedicated Austrian website, including CASUS with the entire RIS plus OGH, VwGH and VfGH case law. No international provider such as Harvey or Legora names Austrian legal sources.

May an Austrian law firm use cloud-based AI at all?

Yes. Section 40 para. 3 RL-BA 2015 expressly permits engaging external providers for electronic data processing, without client consent, where five conditions are met: protected client interests, careful selection, a contractual duty on the provider to report a search of the premises, technical measures reflecting the state of the art, and information to the client about the categories of provider engaged.

What does legal AI cost in Austria?

The range is wide. Publicly visible are RechtGPT at EUR 29 and 49 per user and month with a query allowance, AI:ssociate at EUR 39 to 99 with capped usage, RIDA Online from EUR 89, BEAMON AI at EUR 99 to 199 billed annually, CASUS at EUR 155, or EUR 130 billed annually, and Libra at EUR 900 to 2,400 per user and year. Harvey is reported at roughly USD 1,200 to 2,000 per seat and month with a minimum commitment, which is not a list price. The publisher products name none at all.

Does legal AI replace the legal database?

No, at least not where commentary literature is needed. MANZ and LexisNexis hold content no AI provider can reproduce. The sensible split is by job: the database answers the legal question, the AI tool handles the document work that follows.

Must documents be anonymised before upload?

That depends on the provider and is the most frequent question in first conversations. Where zero data retention is contractually agreed and processing takes place in Switzerland or the EU, prior anonymisation is not required. What should be checked is whether the provider has ruled out caching and human review at its model suppliers. At CASUS client content sits in Zurich, inference runs in Belgium, Switzerland and Sweden, and zero data retention is agreed, so prior anonymisation is not needed.

How should the ÖRAK offers be judged?

The ÖRAK promotes two: AI:ssociate with three free months and BEAMON AI with six free months plus a 20 % standing discount. Both are an inexpensive entry point. Both state Austrian primary sources: AI:ssociate statutes and supreme court case law, BEAMON AI a RIS integration. Before deciding, the follow-on price and the allowances should be settled, and the trial should run on real files rather than sample documents.

What matters in a Word integration?

Whether changes are applied to the document in the correct format instead of travelling through the clipboard. An add-in that respects numbering, structure and formatting removes rework; a chat window next to the document does not. Not every provider with a Microsoft connection has a genuine add-in: RechtGPT, for instance, works through a Microsoft 365 Copilot agent.