Legal AI in law firms: why confidentiality comes before features

Featured image: legal AI in law firms, confidentiality before features
More and more law firms are choosing an AI platform, and the vendors' features are converging. What makes the difference is the conditions: where is the data, who sees what, and what does the tool really deliver in daily work? An overview for law firms and legal departments in Germany.

In autumn 2026, hardly a week goes by without another large law firm announcing its choice of an AI platform. The number of large firms that have not yet committed is shrinking noticeably, and billions are flowing into the legal AI market.

For law firms and legal departments in Germany, the question is no longer whether to use AI. The question is: under which conditions is its use compatible with professional rules, client confidentiality and clients’ expectations, and where does it actually deliver?

Key points at a glance
  • The large platforms no longer advertise only better answers but control: information barriers, vetted connectors, verifiable answers.
  • The biggest benefit of AI in a law firm is efficiency: searching, summarising, comparing. The legal assessment and the final document stay with the lawyer.
  • For German law, it matters which sources an AI answers from. Your own files, official sources and legal commentaries are three different things.
  • Data sovereignty belongs at the start of the selection: § 43a(2) BRAO, § 203 StGB and § 43e BRAO also apply to data in an AI system.
  • AI adoption succeeds where partners use the tool themselves and it is clear who reviews which result.

The market is sorting itself by trust

Read the product announcements of recent weeks and a pattern emerges. The large platforms no longer just promise smarter answers. They promise control:

Access

Information barriers

Ethical walls from document management are carried over into the AI environment instead of building a second permission system.

Integration

Vetted connectors

Connections to document management, data rooms, email and chat are advertised with a security review per connector.

Evidence

Verifiable answers

Statements are linked to the exact passage so that invented or outdated citations stand out.

This is no coincidence. The deeper AI reaches into client work, the more it becomes a question of information security. A system that can access the entire file base must respect the same boundaries as every associate and every partner.

Efficiency is the lever, not the finished document

Before turning to security, it is worth taking an honest look at the benefit. Many discussions about legal AI revolve around whether an AI can deliver a legally sound brief or a finished contract. That is the wrong benchmark. No tool takes over your responsibility for a document that leaves your firm, and none should pretend to.

The real gain lies elsewhere: in the groundwork that precedes every legal assessment and takes up most of the hours in many matters.

What AI handles well

  • Searching and summarising files, briefs and correspondence
  • Finding clauses, deadlines and amounts across many documents
  • Comparing versions and documents
  • Providing first drafts and outlines as a starting point

What stays with the lawyer

  • The legal assessment and the strategy
  • Deciding what a finding means for the matter
  • Approving every document that leaves the firm
  • Responsibility towards client and bar

A good tool is therefore one that does this groundwork quickly and backs every result so that the review stays short. A tool that delivers a finished result without a source saves little: either you trust it, or you read everything anyway. And it must be connected to the systems the firm already works in: anyone who first has to export and upload every document loses the time gained straight away.

German law: which sources does the AI answer from?

In the German-speaking market, people now say openly what many firms have long known: legal AI must fit German law, German legal language and German ways of working. A model trained and tested mainly on common law material is not a reliable basis for a question under the German Civil Code.

It pays to look closely at which sources a tool actually answers from. There are three different levels:

SourceWhat it provides
Your own documentsFiles, contracts, briefs, data rooms. This is where most of the work happens and where efficiency gains are greatest, provided every answer points to the passage in the document.
Official sourcesStatutes and court decisions. They make an answer verifiable against the law, if the tool cites the source precisely.
Commentaries and legal literatureThe analysis by expert authors. This content is licensed and only available in tools that hold a licence from the respective legal publisher.

When selecting a tool, ask which of these levels it really covers, and do not settle for the label “legal AI”. An AI that talks about case law without a connected source reconstructs it from memory. That is exactly where invented case numbers come from.

RelatedAI contract review: what AI-assisted contract analysis can do and where its limits are

Data sovereignty belongs at the start

The second point weighs even more: data sovereignty belongs at the start of the selection, not at the end. For professionals bound by secrecy, this is not a matter of style. The duty of confidentiality under § 43a(2) of the German Federal Lawyers’ Act (BRAO) and criminal liability under § 203 of the German Criminal Code (StGB) also apply when client data is processed in an AI system. Firms that involve IT service providers must meet the requirements of § 43e BRAO, including a contractual confidentiality obligation and a check whether a provider abroad offers a comparable level of protection.

In practice, this means clarifying these questions before any feature demo:

  1. Where is client data stored and processed, and under which jurisdiction?Storage location and place of AI processing are two different questions. And a European server owned by a US group falls under the US CLOUD Act.
  2. Are inputs or documents used to train models?The answer should be no, set out in the contract.
  3. How are permissions and information barriers enforced?Anyone not working on the matter should not see its documents through the AI either.
  4. Can every answer be traced to its source?Document, section, verbatim passage. Otherwise the review takes as long as the work itself.
  5. Who is liable if something goes wrong, and what does the contract say?Ask to see the contract under § 43e BRAO and the data processing agreement before you test.
Further readingSection 203 StGB and AI in legal practice: what lawyers really need to know

Legal uncertainty slows things down, clarity speeds them up

A recent Eurobarometer survey by the European Commission among SMEs and mid-caps shows how much this weighs:

42 %already use AI or algorithmic management, a further 21 % plan to
49 %cite legal uncertainty as an obstacle
50 %cite investment costs

For law firms, this is a twofold message. First, clients experience the same uncertainty and expect guidance from their firm. Second, it applies to the firm itself: a firm that has settled the data protection and professional law questions once can roll out faster than a competitor that has to approve every use individually.

The EU AI Act fits this picture. With the Digital Omnibus on AI, approved by the European Parliament on 16 June 2026 and by the Council at the end of June 2026, the obligations for stand-alone high-risk systems under Annex III, which include certain AI systems in the administration of justice, were postponed from August 2026 to 2 December 2027. For AI in regulated products, the date is now 2 August 2028. Most law firm applications are not covered anyway. The breathing space should nevertheless be used to build policies, documentation and training, not as a reason to wait.

RelatedEU AI Act checklist: what companies should check now

Adoption starts with the partners

Technology alone delivers no benefit. A legal AI vendor recently analysed its own usage data: partners now make up 23 % of law firm users there, more than four times as many as a year earlier. Where partner usage is in the top quartile, associates also use the tool considerably more often, by 25 percentage points.

These are vendor figures. But the direction matches the experience from many rollout projects: AI adoption is driven from the top. When partners use a tool themselves, accept work results from it and say so openly, teams follow. If AI is seen as a matter for associates, it remains a pilot project.

Step 1

Choose noticeable use cases

Tasks that partners notice in their daily work: summarising facts, searching files, comparing contracts, extracting deadlines.

Step 2

Define review standards

No AI result leaves the firm unreviewed. The firm defines which review is appropriate in each case, before the start.

Step 3

Make successes visible

Hours saved, faster responses, happier clients. What is measured and talked about gets used.

How aiuno implements this

aiuno was developed together with German law firms and starts where most time is lost: working with your own documents. The legal assessment stays with the lawyer, and that is exactly why every answer is built to be checked quickly.

QuestionHow aiuno handles it
What is aiuno for?For efficiency in daily work: searching files and data rooms, summarising and comparing documents, reviewing contracts in a structured way. aiuno delivers drafts and sources, not legal assessments and not finished documents.
Where do the answers come from?From your own documents, and on request also from official legal sources: statutes and court decisions for most EU countries. For every answer, aiuno names the document, section and verbatim passage and checks before answering whether the quote is really there. aiuno does not connect commentaries or legal literature.
Where is the data?Hosting at Hetzner in Falkenstein, AI processing at AKI.IO in Germany or Scaleway in the EU. No US or non-EU processors for user content. For matters with special requirements, there is a profile with AI processing exclusively in Germany. This also applies to content from connected systems.
How do documents get in?Through standard connectors to around 70 systems, including SharePoint, Exchange and Google Drive, plus API and MCP for specialised systems. Content is mirrored into aiuno through a secured, encrypted interface and kept up to date continuously: what you add in SharePoint also appears in aiuno. You can take over permissions from the source system or manage them through workspaces in aiuno. All integrations
Training and retentionNo training on your data, zero data retention at the AI provider.
Who sees what?Roles per workspace: anyone not authorised for a workspace cannot see its documents through the AI either. Plus two-factor login for everyone and an audit trail. In the profile for special requirements, aiuno also logs who opened which document.
ContractData processing agreement aligned with the minimum content under § 43e(3) BRAO; data processing and § 203 agreements are in place with all service providers. All parties involved are named in the Trust Center.
On request

Statutes and court decisions alongside

On request, aiuno also connects official legal sources: statutes and court decisions for most EU countries, with sources down to the paragraph number. Then the statute or decision that matters sits next to the passage in your file, and just as verifiable.

Conclusion: confidentiality is the foundation, efficiency the benefit

The legal AI market is consolidating, and the features of the large platforms are increasingly alike. What makes the difference is therefore not the features but the conditions: where is the data, who controls access, and can every result be defended towards client and bar?

Firms that answer these questions first roll out AI faster and with less friction. And firms that use AI where it really carries weight, in the groundwork before the legal assessment, gain exactly the time that is missing for the assessment itself.

Note: This article provides an overview and does not replace legal advice in individual cases.

Frequently asked questions

What is legal AI?

Legal AI refers to AI tools built for legal work: searching files, summarising documents, reviewing contracts, preparing drafts. What matters is which sources they answer from and whether every answer is backed by evidence.

Can AI create legally sound documents?

No, and it should not promise to. AI can deliver drafts and sources and take a lot of groundwork off your hands. The legal assessment and the approval of a document stay with the lawyer.

May a law firm put client data into an AI system?

Yes, if the requirements of § 43a(2) BRAO, § 203 StGB and § 43e BRAO are met: careful selection, a contract with a confidentiality obligation and a comparable level of protection for providers abroad.

Does an AI know German commentaries and case law?

Only if these sources are connected. Commentaries are licensed and only available through legal publishers. Without a connected source, an AI reconstructs case law from memory, and that is exactly where invented citations come from.

Is AI in a law firm covered by the high-risk rules of the EU AI Act?

In most cases, no. Annex III covers certain AI systems in the administration of justice; typical law firm applications such as searching your own files or contract review are generally not included. An internal policy and training are still useful.

How does a successful rollout in a law firm work?

With concrete use cases that partners notice themselves, clear review standards before the start and visible successes. When partners use the tool, teams follow.

Hinweis: Dieser Artikel wurde von einem menschlichen Autor verfasst und mithilfe von KI-Tools sprachlich optimiert.

Note: This article was written by a human author and linguistically refined with the help of AI tools.

Uwaga: Ten artykuł został napisany przez autora-człowieka, a jego warstwa językowa została dopracowana przy pomocy narzędzi AI.

Liked this post? Share with others!

Demo

Schenken Sie uns 20 Minuten.

Bringen Sie Ihren größten Datenraum mit. Wir zeigen Ihnen das Produkt. Gerne mit Ihren Testdaten.

Hosting in Deutschland · § 203 StGB-konform · Zero Data Retention

aiuno Screen

Demo

Give us 20 minutes.

Bring your largest data room. We’ll show you the product – happy to use your test data.

Hosted in Germany · No US transfer · Zero Data Retention

aiuno Screen

Demo

Dajcie nam 20 minut.

Przynieście swój największy data room. Pokażemy produkt – chętnie na Waszych danych testowych.

Hosting w Niemczech · Bez transferu do USA · Zero Data Retention

aiuno Screen
Scroll to Top