AI in litigation: tools, risks and opportunities

Litigation is an area where mistakes get expensive. Deadlines, briefs, mountains of documents, and in the middle of it all the question of what AI can and can’t do here. The answer is more nuanced than many expect. AI can take real work off your hands in litigation. But it also brings risks that become a problem without the right safeguards.

This article shows where AI is being used sensibly in litigation today, where its limits are, and what can be learned from real German court decisions.

The basic problem: general-purpose AI tools don’t know legal sources

Many lawyers use ChatGPT or similar tools for first drafts, summaries or research. That’s understandable: they’re fast, accessible and often usable for simple tasks.

But general-purpose AI tools weren’t built for legal work. They aren’t tailored to any specific industry, don’t know legal databases, have no access to recent judgments and can’t tell whether a case citation exists or not. They generate plausible-sounding text, not verified legal content.

In its order of 29 April 2025 (ref. 5 U 1/25), the Higher Regional Court (OLG) of Celle noted that a lawyer had cited four decisions from various higher regional courts that couldn’t be found in any legal database. Shortly afterwards, the Local Court (Amtsgericht) of Cologne reached the same finding (ref. 312 F 130/25): all the sources in a brief had been made up. In both cases the courts assumed the citations were AI-generated; exactly which tool was used has not been established. What both cases have in common: the tool used gave no warning.

What AI can do in litigation

Used correctly, AI can save a considerable amount of time in litigation. The most useful areas:

• Document analysis: large volumes of correspondence, contracts and internal records can be searched and structured with AI in a fraction of the time
• Structuring the facts: building chronologies, analyses of contradictions and overviews of the parties from unsorted files
• Drafting briefs: creating first outlines and wording for statements of claim, grounds of appeal or replies
• Summaries: condensing long judgments, expert reports or witness statements to prepare for hearings or internal discussions
• Research: a first assessment of the legal position and identification of relevant decisions as a starting point for your own analysis

All these areas have one thing in common: AI does the groundwork. The legal assessment, the responsibility and the end product remain with the lawyer.

Three risks that weigh especially heavily in litigation

The following risks aren’t limited to litigation, but the consequences are particularly serious here.

Incorrect source references top the list. No other area depends so directly on correct legal citations as litigation. An invented judgment in a brief damages credibility with the court, can have procedural consequences and can lead to professional disciplinary consequences. The German orders from 2025 document exactly this scenario.

Contradictory submissions are another problem. In the same order, the OLG Celle found that parts of the brief directly contradicted the party’s own submissions elsewhere and were therefore procedurally irrelevant. AI-generated text can be internally contradictory without this being obvious at first glance. Adopting output like this without checking it leads straight into this trap.

Then there’s the question of confidentiality. Litigation documents are among the most sensitive documents there are: client strategies, witness statements, internal communications. Entering them into public AI tools risks breaching professional secrecy and the GDPR. That’s not a theoretical risk; it’s a structural weakness of most common AI tools.

What responsible use of AI in litigation looks like

The right approach to AI in litigation isn’t a ban; it’s a clear framework.

• Check every legal reference and case citation in AI output against a recognised legal database before using it
• Treat AI-drafted briefs as working material, not finished text. Every legal assertion requires a lawyer’s review
• Check all AI-generated submissions for internal contradictions before filing
• Use only data-protection-certified tools for content involving clients

The OLG Celle put it clearly: professional obligations apply in full, regardless of whether AI was involved. Responsibility lies with the lawyer.

Which tools are suitable for litigation

For professional use of AI in litigation, a simple chat interface isn’t enough. You need tools with source verification, structured output and the level of data protection that matters of this kind require.

BEAMON is designed for exactly this. The Extract feature lets you pull specific data points out of large volumes of documents, which is relevant for discovery, due diligence and file analysis. The Assist feature supports research, analysis and drafting briefs with active source verification, grounded in legal sources from more than 20 countries. The Word Plug-in enables review and redlining directly in Microsoft Word, without switching between tools.

What further sets BEAMON apart from general-purpose AI tools: it can be combined with BRYTER Workflows, a no-code platform for automating legal processes. This lets law firms and legal departments build specialised workflows tailored to how they actually work. Examples include

• A claims-handling tool for commercial claims and insolvency cases,
• A file-to-timeline creator that automatically turns files and documents into a chronology,
• A court documents handler for the structured extraction and preparation of court documents,
• A contract reviewer for reviewing contracts, or
• Automating statements of defence in mass litigation.

If you have recurring tasks in litigation, this combination lets you turn them into a structured, auditable workflow instead of tackling them by hand every time. If you’d like to see what that looks like in practice, you can try BEAMON for free: ai.bryter.io/sign-up.

AI in the German justice system: what’s already happening

On the courts’ side too, AI in Germany has long stopped being a topic for the future. The best-known example is OLGA (OberLandesgerichts-Assistent, the Higher Regional Court Assistant), which has been in regular operation at the OLG Stuttgart since November 2022. Developed together with IBM Watson Discovery, OLGA automatically categorises appeal briefs in diesel cases (vehicle model, engine type, emissions standard, recall status) so that similar cases can be handled together. Because of the diesel scandal, the OLG Stuttgart’s caseload jumped from around 1,500 to 15,000 cases a year. Without OLGA, that backlog couldn’t have been cleared. Processing effort is said to have fallen by around half since then.

In Frankfurt, two further projects are in the pilot phase. FRAUKE (FRAnkfurter Urteils-Konfigurator, Elektronisch, the Frankfurt electronic judgment configurator) was developed for the high volume of air passenger rights cases: the system extracts metadata from briefs and creates first draft judgments using text modules. FRIDA, meanwhile, was developed in-house by a judge at the Frankfurt Local Court for regulatory offence proceedings. Both projects are technically promising but not yet fully rolled out. In April 2025, the E-Justice Council adopted an AI strategy for the German justice system, setting the institutional framework for expanding projects like these.

Frequently asked questions