Artificial intelligence has quickly become part of legal practice. Lawyers use it. Clients use it. Courts are confronting it.
But one lesson is becoming increasingly clear: AI can sound remarkably confident while being completely wrong.
Courts across the country have now encountered briefs containing nonexistent cases, fabricated quotations, and authorities that do not actually stand for the propositions for which they were cited. Lawyers have been sanctioned. Courts have referred attorneys to disciplinary authorities. In some cases, the consequences have gone well beyond a monetary penalty.
The problem is not limited to lawyers. Judges and court personnel have also faced scrutiny over AI generated errors appearing in judicial decisions.
AI can be an extraordinarily useful tool. But confidence should not be mistaken for accuracy.
A Case Citation Is Not Real Just Because AI Says It Is
One of the most dangerous characteristics of generative AI is its ability to produce an answer that looks authoritative.
Ask a legal question and the response may contain case names, quotations, statutes, and detailed explanations. The presentation can be convincing enough that the reader assumes the underlying research must exist.
Sometimes it does not.
California courts have already sanctioned attorneys for submitting fictitious or inaccurate authorities generated through AI. Federal courts have likewise imposed serious consequences on attorneys whose filings contained nonexistent cases, misattributed quotations, or significant misrepresentations of actual decisions.
The lesson from these cases is straightforward.
AI may assist with research, but the lawyer signing the filing remains responsible for what goes into court.
AI Usually Knows Less About Your Case Than Your Lawyer Does
There is an important difference between asking AI a legal question and actually litigating a case.
A litigation attorney is not analyzing a hypothetical paragraph.
The attorney is working with the pleadings, evidence, discovery responses, deposition testimony, procedural history, applicable law, opposing counsel’s strategy, the tendencies of the particular court, and the practical economics of continuing the litigation.
AI generally receives only the facts provided by the person asking the question.
And those facts matter enormously.
Change one fact, omit an unfavorable document, misunderstand a procedural event, or describe the opposing party’s position inaccurately, and the resulting analysis can change completely.
This distinction is becoming increasingly important as clients use AI to independently analyze their own cases.
A client may receive an AI generated answer stating that a particular motion should succeed, that a certain claim is viable, or that a particular litigation strategy should be pursued. Because the response is detailed and confidently written, it can appear authoritative.
But the AI may not have reviewed the actual pleadings. It may not know what evidence has been produced in discovery. It may not understand what occurred at a prior hearing. It may not know whether an argument has been preserved, whether a deadline has passed, or whether the factual record actually supports the legal theory it proposes.
Even when AI accurately states a general legal principle, applying that principle to an actual lawsuit requires something different.
It requires judgment.
AI Is a Tool, Not a Second Lawyer
Clients should ask questions, challenge assumptions, and understand the strategy being recommended to them. Lawyers should be prepared to explain why they recommend a particular course.
But there is a difference between using AI to become informed and treating an AI generated response as an independent legal opinion that somehow outweighs counsel’s professional judgment.
That distinction matters particularly in litigation.
A lawyer’s recommendation may reflect weaknesses in the evidence that are not apparent from the client’s description of the case. It may account for procedural limitations, litigation costs, credibility issues, prior rulings, or risks that do not appear in a generalized AI analysis.
The irony is difficult to ignore.
Courts are sanctioning lawyers because they relied too heavily on AI without independently verifying its work. At the same time, attorneys increasingly encounter situations in which AI generated analysis is treated as though it should override the judgment of counsel who actually knows the record and is responsible for litigating the case.
Both situations reflect the same fundamental problem.
They confuse the appearance of authority with actual legal judgment.
Professional Judgment Still Matters
Artificial intelligence will continue to become more sophisticated. Its role in legal research and legal services will undoubtedly continue to grow.
But litigation is not simply an exercise in finding a legal rule and applying it to a paragraph of facts.
Cases develop over months or years. Evidence changes. Witnesses testify differently than expected. Judges make rulings. Discovery reveals facts that neither side knew when the complaint was filed. Strategic decisions often involve choosing among several imperfect options rather than identifying one theoretically correct answer.
AI can process information remarkably quickly.
What it cannot do simply by producing a persuasive answer is transform incomplete information into complete understanding.
The recent wave of sanctions involving fabricated AI authorities is a useful reminder of that limitation. If lawyers themselves cannot safely accept an AI generated case citation without verifying it, litigants should be equally cautious about assuming that an AI generated assessment of an entire lawsuit is necessarily correct.
Technology changes.
The responsibility to exercise judgment does not.
Related practice areas: Civil Litigation · Business and Commercial Litigation