How AI Is Changing Legal Work: Tools and Risks

Law is, at its core, a profession of language: reading, writing, analyzing, and arguing with words. That is precisely what large language models are built to do, which is why legal work is among the professions most directly affected by AI.

Much of a lawyer's day is text-heavy and time-intensive: sifting through case law, drafting contracts and briefs, reviewing thousands of documents in discovery, and summarizing lengthy filings. These are exactly the tasks where AI can compress hours into minutes, reading and generating legal language at a speed no human can match.

But law is also a profession where being wrong has severe consequences. A mistaken citation, a missed clause, or a leaked confidence can lose a case, breach a duty, or end a career. This combination, enormous potential to help and enormous cost of error, makes legal AI a high-stakes topic that must be understood clearly, not adopted blindly.

This cursus is a practical guide for lawyers and legal professionals: what the tools are and do (this lesson), how to use them for the core legal workflows (lesson two), and how to use them responsibly given the profession's strict duties (lesson three). The goal is neither hype nor fear, but a clear-eyed understanding of where AI genuinely helps a lawyer and where it can seriously harm one.

Legal AI is not one thing. It divides into several categories, each targeting a different part of legal work. Knowing the map helps you match a tool to a task.

• Legal research assistants: tools that search case law and statutes and answer legal questions in natural language, often citing sources, replacing hours of manual searching.
• Drafting assistants: tools that generate first drafts of contracts, clauses, letters, and briefs from a prompt or template, which the lawyer then refines.
• Contract review and analysis: tools that read a contract and flag risks, missing clauses, unusual terms, or deviations from a standard, at speed across many documents.
• Document review and e-discovery: tools that sort, classify, and surface relevant documents from the huge volumes involved in litigation.
• Summarization: tools that condense long depositions, filings, or case files into digestible summaries.

The unifying theme is that AI handles the high-volume, language-heavy, first-pass work, reading, sorting, drafting, summarizing, so the lawyer can spend their time on judgment: strategy, analysis, advocacy, and the decisions that require legal expertise.

The crucial framing is that these tools produce a starting point, not a finished product. A research answer must be checked, a draft must be revised, a flagged risk must be evaluated. The lawyer remains the professional; the AI is a fast, tireless, but fallible assistant whose work always passes through human review before it counts.

A practical distinction shapes which tool a lawyer should reach for: general-purpose AI versus law-specific AI. They differ in ways that matter enormously for legal work.

General-purpose assistants are the broad chatbots trained on the open internet. They are versatile and useful for general drafting, brainstorming, and summarizing text you paste in. But they have two serious weaknesses for law: they can invent legal facts and citations that sound authoritative but do not exist, and they were not built with legal accuracy or confidentiality guarantees in mind.

Law-specific tools are built for the profession, typically by connecting the AI to a trusted, curated database of actual case law and statutes. Instead of generating citations from memory, they retrieve answers from real legal sources and cite them, an approach that grounds the output in verifiable material. They are also built with legal workflows and confidentiality in mind.

The key lesson: for anything requiring accurate legal authority, a tool grounded in a real legal database is far safer than a general chatbot answering from memory. But, as the next step warns, even grounded tools require verification, because no legal AI is trustworthy enough to use without checking.

The single most important risk in legal AI has a name: hallucination, when an AI generates false information stated with complete confidence. In most fields this is an annoyance; in law it is a professional catastrophe.

The danger is specific and severe: AI can invent case citations that do not exist, complete with plausible names, dates, courts, and quotations, all fabricated. Because the invented case looks exactly like a real one, a lawyer who does not check may cite a case that was never decided.

This is not hypothetical. In a widely reported 2023 US case, Mata v. Avianca, lawyers submitted a court brief containing multiple fake cases generated by a general AI chatbot, and were sanctioned by the judge once the fabrications were discovered. Similar incidents have recurred since, and courts have responded with penalties and standing orders requiring disclosure of AI use. Filing invented authority can mean sanctions, professional discipline, and severe reputational damage.

Why does this happen? A general language model generates text that is statistically plausible, not verified true. A citation that fits the pattern of a real citation is exactly what it produces, whether or not the case exists. The model has no built-in sense of what is real.

The unavoidable consequence for lawyers is a rule with no exceptions: every AI-provided citation, quote, and legal proposition must be independently verified against a real source before it is relied upon or filed. Law-specific tools that retrieve from real databases reduce this risk substantially, but the professional duty to verify never transfers to the machine. In law, trust-but-verify is really just verify.

The second great risk is not about accuracy but about secrecy. Lawyers hold a strict duty of confidentiality to their clients, and AI tools can quietly threaten it in ways that are easy to overlook.

The core concern is what happens to the information you type in. With many general-purpose AI services, the text you enter may be transmitted to and stored by the provider, and in some cases used to train future models. If a lawyer pastes confidential client information, privileged documents, deal terms, personal data, into such a tool, they may be disclosing client secrets to a third party, potentially breaching their professional duty and, depending on the tool, losing control of that information entirely.

This makes tool choice a confidentiality decision, not just a convenience one. Key questions before using any AI tool with client information:

• Does the provider store what I enter, and for how long?
• Is my input used to train their models?
• Is the data covered by appropriate confidentiality and security terms?
• Is there an enterprise or legal-grade version with stronger protections?

Many law-specific and enterprise tools address this explicitly, offering contractual guarantees that data is not stored or used for training, and appropriate security. General consumer tools often do not.

The practical rule that follows: never enter confidential client information into an AI tool whose data handling you have not verified. Confidentiality is a duty that AI convenience cannot override, and this concern, alongside hallucination, is why the final lesson treats responsible use as inseparable from the tools themselves.

Step back from the individual tools and risks to a balanced view of what AI actually changes for a lawyer, because both the hype and the fear miss the reality.

What AI genuinely changes: the speed and cost of the language-heavy groundwork. Research that took hours can take minutes; a first draft that took an afternoon can appear in seconds; thousands of documents can be triaged rapidly. This is a real productivity shift that frees lawyers from a great deal of tedious work and can make legal help faster and more affordable.

What AI does not change: the lawyer's core professional role. The judgment about what a case means, the strategy for a client, the advocacy in a negotiation or courtroom, the ethical duties, and the ultimate responsibility for the work all remain human. AI does not practice law; it assists someone who does. Critically, accountability does not transfer: if an AI tool produces an error, the lawyer who relied on it is responsible, not the software.

So the realistic picture is neither "AI will replace lawyers" nor "AI is too risky to use." It is that AI is becoming a powerful assistant that changes how legal work gets done while leaving who is responsible firmly with the lawyer. The professionals who benefit most are those who use it to handle the groundwork faster and rigorously verify and own the results.

That balance, real leverage paired with undelegable responsibility, is the theme running through this whole cursus, and it sets up the practical workflows and the ethical rules in the lessons that follow.

AI tools handle the language-heavy groundwork of legal work, research, drafting, review, summarizing, but their output passes through mandatory human verification for accuracy and confidentiality before it becomes the lawyer's own work product.