When AI Overwhelms the Courts: What Businesses Must Learn from the Legal Filing Chaos
The Flood Begins: When Artificial Intelligence Pushes Courts to Their Limits
It sounds at first like a problem that exclusively affects lawyers, judges, and legal policymakers: Germany's courts are increasingly being inundated with AI-generated briefs. Submissions hundreds of pages long, created in minutes, often inadequately substantiated or riddled with fabricated citations, are threatening to drive the justice system into a structural crisis. The Federal Ministry of Justice is reportedly working "at full speed" on a reform of procedural rules, CSU lawyers are demanding "equality of arms" between the legal profession and the judiciary, and politicians across party lines are recognizing the seriousness of the situation.
Yet anyone who looks closely will recognize: the judiciary is no special case here. It is a magnifying glass that reveals what happens with AI-generated content in any organization when structure, review, and accountability are absent. The mistakes that are currently driving judges to despair occur daily in businesses as well – in internal reports, proposals, customer communications, and decision-making documents.
From the Bottleneck of Writing to the Bottleneck of Reading
The fundamental problem is deceptively simple: AI makes the generation of text nearly costless. What once took hours now takes seconds. This sounds like pure efficiency gains – and it is, as long as what is produced can be reviewed, contextualized, and accounted for.
This is precisely where the blind spot of many implementation strategies lies. The bottleneck shifts. In the past, writing was expensive, reading comparatively quick. Today, writing is almost free, but the careful reading, reviewing, and contextualizing of content remains a human, time-intensive effort. Anyone who ignores this connection and simply layers AI on top produces more – but not better.
"AI scales whatever you give it: diligence just as much as carelessness. The value is created through structure, review, and accountability – not through the sheer volume of text generated."
Dr. Maik Bunzel, founder and managing director of mabucon.eu, who supports companies in building tailored AI agents and automation workflows, puts it with precision: organizations that adopt AI without simultaneously defining who reviews and approves output are not creating efficiency – they are creating uncontrolled output. The difference between the two is not academic, but directly relevant to business outcomes.
Hallucinations: When AI Errs with Confidence
A particularly striking example is provided by a ruling of the Kammergericht Berlin from November 2025. A lawyer had incorporated court decisions as citations into a brief that had simply been fabricated by an AI – legally "hallucinated." The Kammergericht responded with a sharp reprimand. The reputational damage was real, the legal consequences looming.
What leads to a high-profile ruling in the justice system happens daily behind closed doors in companies: AI systems generate content that is factually wrong with great conviction. Fabricated statistics in a presentation. Non-existent studies in a strategy paper. Incorrect product specifications in a customer proposal. The consequences vary in severity depending on the context, but the pattern is identical.
Hallucinations are not a fixable bug that the next model version will fully resolve. They are a structural characteristic of current Large Language Models. The right response is not to avoid using AI – but to design processes so that critical outputs are verified before being passed on. This means: defined approval levels, clear responsibilities, and, where possible, automated cross-checks by specialized agents.
AI reinforces – it does not sort
Another often underestimated phenomenon also emerges in the justice debate: AI systems deliver arguments, not assessments. Ask a language model for reasons in favor of a particular position, and you get compelling reasons – regardless of how strong those reasons actually are. Ask for counterpositions, and you get equally compelling counterarguments.
In the justice system, this leads to briefs that sound voluminous and confident but are thin on substance. In companies, it leads to decision papers that are fluently worded and appear professional, yet conceal significant risks or alternatives on the merits. The trust that a well-written text generates becomes a trap when no one critically scrutinizes the content.
The solution lies not in distrust of AI, but in the deliberate design of usage scenarios. Where AI creates drafts, a human or agentic cross-check is needed. Companies introducing AI-supported workflows today should plan for both a creation layer and a filtering layer from the outset.
The other side is arming up – and companies need both
The political response to the justice crisis is telling: the demand is not to ban or restrict AI, but to establish a level playing field. Courts and authorities are to receive their own AI assistance systems to handle the growing volume of input in a structured and efficient manner. The Federal Ministry of Justice, the Union parliamentary group, and representatives of all factions are remarkably united on this point.
This principle applies to companies as well. Anyone with customers, partners, or internal stakeholders who use AI for their communications will increasingly face a changed information density. More inquiries, more documents, more arguments – in less time. Those who deploy AI only on the creation side will fall behind. What is needed are tools that also structure, prioritize, and assess incoming information for relevance.
Dr. Maik Bunzel from mabucon.eu observes in practice that companies frequently underestimate this dual need – AI for creating and AI for filtering. The first wave of automation typically captures content production. The second, more demanding wave concerns the intelligent processing of incoming information. Those who think both together create genuine operational resilience.
Regulations are coming – those who are internally prepared have a head start
In the judiciary, concrete regulatory responses are emerging: page limits for briefs, structural requirements for submissions, transparency obligations regarding AI use, and expanded sanctions for abusive practices. The Federal Ministry of Justice has announced it will present corresponding reform proposals within six months.
For companies, this development is an early signal: sector-specific regulation of AI use in communication – with authorities, in contract matters, in customer communications – is foreseeable. The EU AI Act already establishes a legal framework that will be filled with industry-specific provisions in the coming years.
- Those who document AI use internally and have established approval processes can quickly meet compliance requirements.
- Those who have already defined quality standards for AI output will not need to build processes from scratch when regulatory requirements arrive.
- Those who have clearly defined responsibilities – who reviews, who approves, who is liable – stand on solid ground when external requirements arise.
- Those who anticipate transparency obligations and label AI-generated output as such where necessary build long-term trust.
The Real Opportunity: Access, Speed, Focus
As real as the risks are – the opportunities deserve at least as much attention. In the judiciary debate, even critical voices point out that AI makes it easier for people to access the law who previously could not afford legal advice. Routine processes become faster. Standardisable tasks can be handled more efficiently.
For companies, this means: AI lowers the barrier for high-quality communication, structured analysis, and scalable processes. Small teams can keep up with the output of growing teams. Expert knowledge becomes more accessible and applicable. Decisions can be better prepared when AI structures and presents relevant information.
The key does not lie in whether AI is used – that question is no longer relevant for most companies. The key lies in how its use is organised. With what guardrails, what review routines, what responsibilities.
Conclusion: Structure Beats Volume
The judiciary debate is a case study that points beyond the legal system. It shows with rare clarity what happens when technology is introduced into existing processes without accompanying structure: the bottleneck shifts, quality suffers, trust is damaged, and regulation follows.
For companies, the consequence is not to put AI projects on ice. It is to ask the right questions from the outset: Who is responsible for the output? How is it reviewed? What happens when something is wrong? Which processes require human approval, and which can run in an automated fashion?
Dr. Maik Bunzel, founder and managing director of mabucon.eu, summarises the core conviction precisely: AI agents and automation workflows deliver their value not through the volume of output generated, but through the quality of the decisions they prepare and support. Those who understand this are not only better protected against the risks – they harness the full potential of the technology.