The most common mistakes when using AI in legal practice all trace back to one failure: trusting the tool without verifying what it returns. Artificial intelligence can speed up legal drafting and legal research, but it can also invent case citations, leak sensitive data, and miss regulatory nuance. Responsibility for a filing always rests with the attorney, not the system.
What you’ll take away from this
- The most serious mistake when using AI in law is citing nonexistent case law generated by AI without verifying it against reliable legal databases.
- Professional responsibility for a filing always rests with the attorney, never with the AI tool that helped draft it.
- Entering confidential client information into open generative AI tools can breach attorney-client privilege and data protection obligations.
- AI in legal practice works as a support tool, but legal judgment and human review remain the responsibility of the legal professional.
- A clear internal policy on AI use reduces the risk of errors at any law firm.
If you work at a law firm, you’ve already seen this scene play out. A colleague pastes in an AI-generated draft, cites three cases that sound perfect, and files the brief. Two of those cases don’t exist. The cause is a missing verification process. This article walks through the five most common mistakes when using AI in legal practice and explains how to avoid them with professional judgment, human oversight, and an internal policy that protects your firm and your clients.
What AI mistakes in legal practice are, and who they affect
AI mistakes in legal practice are failures that generative AI makes in legal contexts: inventing case law, exposing confidential data, applying outdated rules, or missing nuances that only an attorney would catch. They affect lawyers, law firms, and legal professionals who adopt these tools without judgment or controls in place.
The legal sector is especially sensitive to this for three reasons. First, a filing with false information can have direct procedural consequences. Second, client confidentiality is protected by attorney-client privilege and by data protection obligations. Third, responsibility always rests with the professional, never with the tool.
Legal AI doesn’t fail out of malice. It fails because it generates plausible text, not verified text. A language model predicts which words fit together; it doesn’t check whether a case was actually decided. According to a 2023 analysis published by Stanford HAI, large language models show factual hallucination rates ranging from 3% to 27% depending on the task, a range that has direct consequences for the validity of legal filings. That’s why legal practice demands a human filter at every step. AI helps you draft, summarize, and research, but legal judgment is still yours. If you want the full picture, this guide to using AI in law walks through the criteria for responsible use step by step.
Citing fake AI-generated case law: the mistake that can cost you the most
The most well-documented failure is citing case law that doesn’t exist. AI generates references with flawless formatting, docket number, court, and date, but the decision never happened. This mistake can cost you credibility with the court and sanctions for bad-faith litigation conduct.
The case that set the global precedent is Mata v. Avianca (2023): two attorneys filed a brief citing six cases invented by ChatGPT and were sanctioned by the judge with a $5,000 fine. According to the court’s own order, none of the citations could be located in official databases. Similar cases have since been documented in other countries, including the United Kingdom and France, according to reporting by Law.com and Le Monde in 2024.
AI can invent citations because it completes patterns. When you ask for case law on a topic, it produces what statistically resembles a real decision, complete with plausible names and references. It doesn’t check an official repository unless it’s connected to one. Without that connection, an AI-generated citation is a hypothesis dressed up as a fact.
The solution is simple: never accept a citation as valid without checking it against an official source.
How to verify case law with human review before filing a brief
Before signing off on any filing, apply a four-step verification process:
- Find the citation in an official source. Look up each case in an official case law database, such as your court’s own reporter system, Westlaw, Lexis, or whatever legal research platform your firm relies on.
- Check the docket number, court, and date. Confirm all three match the real document, not just the case name.
- Read the cited holding. Confirm the decision actually says what the AI claims it says, not something close to it.
- Log the verification. Keep a record of who reviewed each citation and when.
Human review is not optional for AI-generated case law. It’s the last barrier between a useful draft and a sanctionable error.
Exposing confidential client information when using generative AI
Entering sensitive client data into open generative AI tools can breach attorney-client privilege and your data protection obligations. Many free AI platforms use conversations to train their models, which means the information leaves your firm’s control.
Think about what gets pasted into a prompt without a second thought: names, dollar amounts, litigation strategy, internal documents. When you use AI in the public cloud, that data travels to third-party servers. According to their own published terms, several widely used generative AI platforms reserve the right to process submitted content in data centers outside the United States when you haven’t signed up for a business plan with specific contractual safeguards.
This runs directly into two obligations:
- Attorney-client privilege, which protects everything a client shares with counsel.
- Data protection obligations, which require a legitimate basis and safeguards for handling personal information.
The practical rule is simple: never enter identifiable confidential information into a tool that doesn’t offer contractual guarantees on confidentiality and data handling. If you need to use AI with real data, choose closed-environment solutions with data processing agreements and no reuse of your data for training. Knowing your way around the legal AI tools available to attorneys helps you tell open platforms apart from environments with real safeguards. When possible, anonymize before you paste. Misusing these platforms is one of the hardest AI risks to reverse, because once data is exposed, it doesn’t come back.
Blindly trusting AI and outsourcing legal judgment
Outsourcing legal judgment to AI is the third serious mistake. The tool drafts and summarizes, but it doesn’t understand the regulatory context, the client’s history, or the strategy behind the matter. Trusting its answer without review turns the attorney into a mere copyist.
AI carries the biases baked into the data it was trained on. It may favor a majority interpretation and overlook a minority line of case law that actually fits your matter better. It can also apply rules from another jurisdiction or an outdated version of the law, because it doesn’t track the timing nuance unless you specify it precisely. No AI tool knows the law that applies to your specific matter better than you do.
Here’s a concrete example: you ask AI to analyze whether an appeal is viable. It returns a coherent analysis but leaves out a recent legal change that shifted the filing deadline. The text sounds convincing. If you sign it, the mistake is yours.
Professional judgment is exactly what AI doesn’t capture:
- A strategic read of the matter and the specific court involved.
- The weight of circumstances that never make it into the prompt.
- The ethical judgment of which argument is worth making.
Use AI as a starting draft. Legal judgment, the final decision, and responsibility all remain with the attorney.
Using AI without legal training or human oversight
The risk shows up when these tools are used without professional judgment. A powerful tool in the hands of someone without the technical competence to evaluate it produces filings that are fast but dangerous.
Without training, a legal professional can’t tell when an AI tool’s answer is plausible but wrong. They don’t catch the invented citation, the bias, or the outdated fact. Speed becomes an accelerant for errors. According to the International Bar Association’s report AI and the Legal Profession (2023), 64% of lawyers already using generative AI in their work said they had received no formal training on its limitations or on verification protocols.
AI literacy for legal professionals covers three basics:
- How AI tools actually work, so you understand why they hallucinate and where they break down.
- How to write precise prompts, to reduce errors at the source.
- How to verify and oversee the output, to keep professional control in your hands.
Human oversight isn’t a formality. It’s professional responsibility applied to every use of AI. Ongoing training lets you keep pace with technologies that change faster than any manual can. Building a foundation in applied AI gives your team the basis to decide when to trust a tool and when not to. Without that foundation, the legal profession takes on a risk that no tool can offset.
How to bring AI into a law firm with an internal policy
Bringing artificial intelligence into a law firm requires a clear internal policy before the tool itself. Technology without usage rules generates errors; with a defined framework, it becomes a controlled advantage. Adopting AI doesn’t start with buying software. It starts with deciding how it will be used.
An internal AI use policy needs to answer specific questions:
| Area | What the policy must define | Why it matters |
|---|---|---|
| Approved tools | Which AI systems can be used, and which can’t | Prevents the use of platforms without confidentiality guarantees |
| Confidential data | What information never goes into open AI | Protects attorney-client privilege and data protection obligations |
| Verification | Who reviews citations, facts, and case law before filing | Reduces the risk of fake case law |
| Disclosure of use | When AI use is disclosed to the court, if applicable | Reinforces good-faith litigation conduct |
| Recordkeeping | How each relevant use of AI is documented | Supports traceability and accountability |
The American Bar Association and most state bar associations have published guidance on the responsible use of artificial intelligence in legal practice. Using that guidance as a framework, alongside the rules of your own jurisdiction, helps align your firm’s policy with what the profession expects. Adopting AI only creates value when the use of emerging technology comes paired with clear rules and well-defined accountability.
Disclosing AI use to the court deserves special attention. It isn’t always required, but when a tool has played a substantial role in a filing, being transparent about it protects your good-faith standing with the court. It’s also worth telling your clients when AI has played a role in their matter, since that reinforces trust and transparency. Leadership in the responsible use of AI at a firm largely comes down to setting these standards before a conflict arises.
Best practices for the responsible use of AI in legal practice
Turn the policy into a checklist your team can follow every day:
- Mandatory human review of every filing before it goes out.
- Confidentiality controls: never enter identifiable client data into open AI.
- Case law verification against official databases.
- Team training on the responsible use of artificial intelligence.
- Recordkeeping for AI use in matters where its involvement is material.
These five practices cover most of the risk for a reasonable amount of effort, and they help your firm make professional, everyday use of artificial intelligence.
Frequently asked questions about AI mistakes in legal practice
Who is legally responsible if AI makes a mistake in a legal filing?
The attorney who signs the filing. Professional responsibility rests with the lawyer, not with the AI tool that helped draft it. AI is a support tool, just like any database or template. If a filing cites fake case law, the court holds the professional accountable, since they had a duty to verify before signing.
What mistakes can AI make in legal practice?
AI can invent decisions and citations that don’t exist, apply outdated rules or the wrong jurisdiction’s law, carry biases from its training data, and expose confidential information if it’s entered into open platforms. It can also miss relevant nuances in a case. All of these failures share one cause: AI generates plausible text, not verified text, and it needs human review at every step.
Is it safe to enter confidential client data into generative AI tools?
No, unless you’re using a solution with contractual guarantees on confidentiality and data handling. Many open generative AI tools retain and process what you enter, sometimes even to train their models. This can breach attorney-client privilege and data protection obligations. Anonymize the data or use closed environments with data processing agreements before working with real client information.
Do you have to disclose AI use to the court when drafting a filing?
It isn’t always required, but it’s worth doing when AI played a substantial role. Being transparent about it reinforces good-faith litigation conduct and avoids problems if the court detects AI-generated content. Your firm’s internal policy should set out when and how disclosure happens. When in doubt, transparency protects you more than silence does.
Can AI replace an attorney’s legal judgment?
No. AI drafts, summarizes, and researches, but it doesn’t understand the full regulatory context or the strategy behind a matter. It carries biases and can miss relevant case law or apply the wrong jurisdiction’s law. Legal judgment, ethical evaluation, and the final decision are the attorney’s responsibility. AI is a starting draft, never a substitute for professional judgment.
What should an internal AI use policy at a law firm include?
It should define which tools are approved, what information never goes into open AI, who verifies citations and case law, when AI use is disclosed to the court, and how each relevant use is documented. It’s worth aligning the policy with guidance from the American Bar Association and your state bar. A clear policy reduces the risk of errors across the whole firm.
Your next step with AI in legal practice
AI is useful at a law firm, but only with judgment, human oversight, and training. The five mistakes covered here, from citing fake case law to compromising confidentiality, share the same root cause: using the tool without a verification process and without the competence to judge what it returns. With the right training, that risk goes down, and AI starts saving you time on repetitive tasks.
If you want your team to apply AI with a practical, responsible approach, Founderz’s AI training for law firms works through these exact cases with firms and organizations. Founderz partners with more than 1,700 companies and, together with Microsoft, builds training around real work rather than theory. With more than 700,000 students trained across more than 170 countries, the approach has been tested in demanding professional settings. The first step is deciding how you want AI to enter your legal practice, before it enters on its own.