// AI for Businesses

AI for Lawyers: 5 Pitfalls That Cost Law Firms Time, Money, and Reputation

Prawbot – interfejs AI dla prawników, polskie narzędzie z bazą 4,7 mln dokumentów | jsoncrew.com

8-minute read · By Maciej Ramus

  • A lawyer in New York filed a brief containing six court rulings generated by ChatGPT. None of them actually existed. Fine: $5,000.
  • Popular chatbots are not aware of the 2023 amendments to Polish law and provide outdated legal information.
  • Every query sent to ChatGPT leaves the EU. Customer data is sent to servers in the U.S.
  • A flat-rate subscription fee is charged even when the law compaNos isn't conducting research.
  • Case law research: 3 hours under normal circumstances. With the right tool: 5 minuteses.

More and more law compaNoss are turning to AI. Case law research, drafting pleadings, summarizing documents. It sounds great on paper.

The problem is that Most of the AI tools available on the market were not designed with Polish law in mind. And these differences are very specific: nonexistent court rulings in legal documents, leaks of customer data abroad, and bills for subscriptions you don't use.

Below are five pitfalls that many law compaNoss fall into today. And here's what to do about them.

1. The chatbot gave you a verdict that doesn't exist

This isn't a story from Twitter. In May 2023, New York attorney Steven Schwartz filed a brief in federal court citing six rulings generated by ChatGPT. None of them existed.

The case numbers looked convincing. The parties, dates, and brief descriptions of the rulings—everything was consistent. The court requested an explanation. When it turned out that the judgments were fictitious, a $5,000 fine was imposed. The case was covered by Reuters, The New York Times, and dozens of legal media outlets.

The mechanism behind language models means that Hallucinations are an inherent risk, not an isolated error. The model generates text that „sounds” like a real court ruling, because that’s how next-word prediction works. The reference number, page numbers, date, and brief description—everything is consistent. Everything is made up.

For general-purpose chatbots, this is a flaw. For a lawyer, it poses reputational, disciplinary, and litigation risks.

How it should work: Every legal ruling cited by the AI system should be automatically verified in the Polish case law database before it is included in the document. Not afterward, not manually—automatically, in the background, for every citation.

2. You pay a fixed monthly fee, even when the law compaNos isn't conducting any research

Most AI systems for lawyers are billed under a SaaS model: a fixed monthly fee per user. This is regardless of how many cases required case law research that month.

During months with lower occupancy, vacation periods, or for projects that don't require in-depth research, the invoice amount is the Same as during periods of full utilization.

A specific example: Four lawyers, monthly license fee of 600–80PLN 0. August, vacations, fewer cases. The system was not used for three weeks. Invoice: identical to the one in OCTOber when the system was in full use.

An alternative is the model pay-per-use: You pay only for what you actually use. If the law compaNos didn't conduct any AI research in a given month, that month costs zero zlotys.

3. Customer data is transferred abroad

ChatGPT, Claude, Gemini. The three most popular AI chatbots. Three servers located outside the European Union.

Every query, every document snippet, and every customer name you paste into the chat is sent to servers in the United States. The standard terms of service for OpenAI, Anthropic, and Google cover data processing for the purposes of the service, but are not subject to the requirements of attorney-client privilege or legal counsel-client privilege.

Article 6 of the GDPR requires a legal basis for transferring data outside the European Economic Area. Data Processsing Agreements (DPAs) with vendors are one thing. Whether the use of tools hosted in the U.S. meets the standards for the protection of professional secrecy required by Polish law is another.

A safe alternative: A system operating in a Polish data center or deployed on-premises on the law compaNos’s infrastructure. Client data does not physically leave the country.

Accuracy on legal tasksPL · benchmark
Prawbot94%
Western Giant in Legal AI96%
Generic chatbot61%
A cutting-edge product at a fraction of the price of Western solutions.

4. The chatbot is not familiar with Polish law from the past two years

Lanuage models have a knowledge cutoff date. GPT-4, in its everyday version, has knowledge up to a certain point in 2021–2023, depending on the release. It is not aware of the 2023 amendment to the Code of Civil Procedure, the 2024 changes to tax regulations, or the Supreme Administrative Court’s ruling from last quarter.

You're asking about the current legal situation. You receive an answer based on the situation from a year or two ago. The model won't tell you that its knowledge is out of date. He will answer just as confidently as he would a question to which he knows the current answer.

An AI tool designed for Polish lawyers should be updated on an ongoing basis: judgments from common courts, interpretations from the National Tax Information Service, rulings from the Supreme Administrative Court and Regional Administrative Courts, legal acts from the European Law Journal, and documents from the Polish Financial Supervision Authority.

Not general knowledge about the world. Only Polish law, as it currently stands, with the ability to verify every citation.

5. Case law research: 3 hours instead of 5 minuteses

A typical case law search: five legal websites open in tabs. Manually searching each one individually. Copying case numbers into a document. Verifying that the case numbers are correct. Repeating this process for subsequent rulings. Three hours of work, sometimes more for complex cases.

Case law databases have been digital for a long time. What’s missing is a tool that search them all at once, will return with the results and provide them in a format ready for use in a pleading, with signatures that can be verified immediately.

One search. The entire Polish case law database. Results in minutes. This should be the standard, not a premium feature.

PBLegal Opinion I'm checking the quotes
Analysis of the Facts
Searching for case law and statutes
Verification of reference numbers in the database
Editing of reviews (10–20 pages)
III CZP 45/15Article 471 of the Civil CodeII CSK 293/16

Prawbot: A Polish AI Built Exclusively for the Law

Prawbot It is an AI system designed specifically for the Polish legal community. It is not a general-purpose chatbot adapted to the law. It is a tool built from the ground up with a lawyer’s work in mind.

Real-time signature verification. Every court ruling referenced by the system is verified in the Polish database before it is included in the document. There is zero risk of false rulings appearing in a pleading.

Pay-as-you-go. No fixed monthly fees. Free accounts for your entire law compaNos team. 10PLN 0 in starting credit without a credit card. You pay only for what you actually use.

The data remains in Poland. Infrastructure in a Polish data center. Option to deploy on-premises on the law compaNos’s servers. Client data does not leave the country.

An up-to-date, comprehensive database of Polish laws. 4.7 million documents: judgments from common courts, interpretations from the National Tax Information Service (KIS), rulings from the Supreme Administrative Court (NSA) and Provincial Administrative Courts (WSA), legal acts from the ELI database, and documents from the Polish Financial Supervision Authority (KNF). 14 collections, 217 legal specialties. Not general knowledge—exclusively Polish law.

Research in minutes, not hours. One query, the entire body of Polish case law. Research that normally yeses 3 hours yeses just 5 minuteses with Prawbot.

If your law compaNos is considering implementing AI or evaluating the available options, Prawbot offers a free account with no credit card required and 10PLN 0 in starting credit.

Sign up for a free account at prawbot.pl →

JK
#1 · John Smith, Attorney at Law · 2:32 p.m.
I've uploaded 180 pages of case files; summarize the case and create a timeline.
//
Prawbot 1.1 · 2:33 p.m. · 1,240 cycles
Dispute over payment under a supply contract. The seller is seeking payment, while the buyer is claiming defects and a delay. The amount in dispute is approximately 184,00PLN 0.
12.03.2023 conclusion of a supply contract
20.02.2024 delivery with quantity shortages
05.2024 late payment (45 days)
09.2024 debit memo for 18,40PLN 0
11.2024 demand for payment
⚠ Risks: Statute of limitations on the claim + defense based on setoff against the warranty
✓ extract· 180/180 files · timeline (5 events) · 2 risks
● Every citation checked against the database

Diagnosis: 3 Questions to Ask Before Choosing an AI Tool for Your Law Firm

  1. Where is the data physically stored? Ask directly about the data center and on-premises options.
  2. How does the system verify the cited rulings? Automatic verification of the signature before inserting it into the document is the bare minimum.
  3. What does the pricing model look like for low occupancy and full occupancy? A fixed-rate plan may end up being more expensive than it appears when you sign the contract.
5 out of 5 (11 votes)
// Contact

Schedule a free consultation

Tell us about your project—we'll get back to you with a detailed plan and a quote. No strings attached, just a clear price.

contact@jsoncrew.com