When Legal Work No Longer Requires a Lawyer

Judges are becoming the AI firewalls of the legal system.

They are issuing courtroom rules, demanding disclosures, reprimanding attorneys and sanctioning those who submit fictitious cases produced by generative AI. In one remarkable federal case in Mississippi, lawyers representing both sides submitted filings containing fabricated legal authorities. U.S. District Judge Sharion Aycock disqualified all four attorneys, imposed financial penalties and canceled the approaching trial.

Last week, Chief U.S. District Judge Hala Jarbou warned the federal government after a Justice Department filing in an immigration case cited a decision the court could not find. The supposed Sixth Circuit case appeared on a reporter page for an unrelated commercial arbitration decision. The quotation attributed to the nonexistent case could not be found anywhere else.

“It should be obvious that any attorney who uses AI must scrupulously review its work product,” Jarbou wrote.

She is right. A lawyer cannot outsource the duty of candor to a machine. If your name is on the filing, you own every word, argument and citation in it.

But judges are not standing outside this transformation. They are trying to determine how to use AI themselves.

More than 60% of the 112 federal judges responding to a recent Northwestern University and New York City Bar Association survey said they had used at least one AI tool in their judicial work. About 22% used AI weekly or daily. Legal research was the most common application, followed by document review.

Some judges are going further. They are using AI to generate case timelines, analyze filings, identify weaknesses in arguments, suggest questions for hearings and prepare initial drafts of orders after they have decided how to rule.

That makes the judge both the firewall and a user of what the firewall is supposed to contain.

The risks are not theoretical. In 2025, two federal judges acknowledged that members of their chambers had used generative AI in connection with error-filled judicial documents. One order reportedly included nonexistent parties, a fictitious case and misstatements of state law. Another opinion contained invented quotations and inaccurate descriptions of earlier decisions.

One judge attributed the errors to a clerk who had used Perplexity. The other said a law school intern had used ChatGPT without authorization.

Those explanations should sound familiar. Law firms have offered versions of them after being caught submitting hallucinated authorities: A junior lawyer used the tool. Local counsel signed without reading. Someone violated firm policy. The supervising partner did not know.

Courts have correctly rejected those excuses. The person whose signature appears on a legal document remains responsible for it. The robe cannot create a different standard. When a judge signs an order, the work of the clerk, intern and AI system becomes the work of the court.

As Louisiana Judge Scott Schlegel put it in an Axios interview, “When we sign an order, if there’s something wrong in it, that’s the law.”

That is reason enough for the judiciary to establish approved tools, confidentiality protections, verification requirements, training and clear rules governing AI use in chambers. Judges should not upload sealed records or sensitive personal information to consumer AI systems. They should understand how models retain and process information. They should know when AI has materially shaped an analysis, not merely corrected its grammar.

But the procession of hallucinated cases is not the larger story. It is the most visible symptom of an immature technology entering an institution that moves slowly and carries enormous consequences when it gets something wrong.

The bigger question is what the legal profession will look like when people can obtain legal work without going through a lawyer.

AI is Separating Legal Work From Lawyers

The legal profession has protected its position through licensing, bar admission, ethical rules, unauthorized practice laws, and procedural complexity. Some of those barriers exist for very good reasons. Bad legal advice can cost people their freedom, children, homes, businesses and life savings.

But these protections also preserve a professional monopoly constructed around an assumption: Ordinary people cannot understand, produce or navigate legal work without lawyers.

AI challenges that assumption.

Someone can already upload a lease and ask an AI system to identify unusual provisions. A small-business owner can ask it to explain a contract, compare proposed language with common industry terms and draft revisions. An employee can describe a workplace dispute and receive a preliminary assessment of possible claims. A consumer can generate a demand letter. A self-represented litigant can prepare a first draft of a complaint, response or motion.

The results are not consistently reliable. AI can omit important facts, misunderstand jurisdictional differences and state the law with extraordinary confidence while being completely wrong. Anyone relying on today’s general-purpose systems for consequential legal advice should understand those limitations.

But “AI sometimes hallucinates” is not a durable defense for the legal profession.

The systems will improve. Legal AI will become more tightly grounded in current statutes, regulations, court rules and authoritative case databases. Citation checking will increasingly become automatic. Models will be restricted by jurisdiction, practice area and procedural posture. Different agents will draft, critique, verify and test one another’s work before a document reaches a human.

The legal profession cannot base its future on the hope that the technology will remain as unreliable as it is today.

The current wave of judicial sanctions and AI disclosure rules should be understood as an intermediate step. Courts are effectively saying that AI may assist in producing legal work, but a licensed human must remain responsible for the result.

That is prudent today. It also preserves the lawyer as the mandatory human wrapper around an increasingly automated product.

How long will that remain necessary for every legal service?

Does someone need a lawyer to prepare a routine will, or do they need a legally valid will? Do they need a lawyer to review a residential lease, or do they need to understand the lease’s risks? Do they need a lawyer to write a demand letter, or do they need a credible letter grounded in the applicable law?

Needing legal work is not always the same as needing a lawyer.

Lawyers May Become the New Software Engineers

Software developers offer a preview of what comes next.

Developers are not disappearing because AI can generate code. But their jobs are changing rapidly. Less of their value comes from personally typing every line. They increasingly define requirements, supply context, choose architectures, supervise coding agents, test output, correct failures and accept responsibility for the finished system.

Lawyers are heading toward a similar model.

They will frame legal questions instead of conducting every search. They will direct legal agents instead of drafting every paragraph. They will verify authorities instead of manually assembling every citation. They will concentrate on strategy, negotiation, advocacy, risk and judgment.

A lawyer may manage the production of a motion much as a software engineer manages the production of AI-generated code.

That does not mean the lawyer becomes irrelevant. It means the value moves.

A lawyer who defines the profession primarily as producing legal text faces real disruption. A lawyer who can understand a client, exercise judgment, recognize what the machine missed, construct a strategy, persuade a court and accept responsibility for the outcome may become more valuable.

The profession survives, but its work and economics change radically.

The billable hour is particularly exposed. If a legal agent can produce in minutes what an associate once needed 15 hours to research and draft, clients will eventually stop paying for the eliminated labor. Firms may attempt to preserve fees by emphasizing the value of the result rather than the time invested. That may work for high-stakes matters. It will be much harder to defend across routine research, review and document production.

AI also threatens the traditional law firm pyramid. Firms employ large numbers of junior lawyers to review documents, research questions and prepare drafts that more experienced lawyers refine. That work generates revenue while training the next generation.

What happens when legal agents do most of it?

A firm may require fewer associates. Clients may refuse to pay for junior lawyers learning on their matters. Yet without years spent researching, reviewing and drafting, young lawyers may not develop the judgment needed to supervise AI later in their careers.

Law faces the same paradox emerging in software development: If AI performs the entry-level work, how does someone acquire senior-level expertise?

Access to Justice or Access to Something That Looks Like It?

The greatest opportunity is also the profession’s greatest competitive threat.

Most Americans cannot afford conventional legal representation. Many people go without help or attempt to represent themselves, not because their matters are unimportant but because the cost of hiring a lawyer is disproportionate to what is at stake.

A reliable AI system that charges $50 to review a lease, prepare a basic will, respond to a debt-collection claim or explain an employment dispute would not simply take business from lawyers. It would create a market among millions of people who were never going to hire one.

That could be the most significant expansion of access to legal services in generations.

It could also create a flood of AI-produced complaints, motions and appeals that look authoritative but contain weak arguments, irrelevant precedent or invented facts. Damien Charlotin’s AI Hallucination Cases Database now contains more than 1,700 matters in which courts have addressed suspected or established AI-generated errors. More than 1,000 involve self-represented litigants.

AI could make the courthouse more accessible while simultaneously making it harder for the court to function.

There is also no reason to assume the technology will equalize power automatically. A tenant may have an AI assistant. The property-management company will have an entire legal-agent platform integrated with its records, contracts and litigation history. An employee may use a chatbot. The employer will deploy specialized systems trained on thousands of previous claims.

Giving everyone access to AI is not the same as giving everyone equal access to justice.

Still, the answer cannot be to preserve unaffordable legal services because the alternative carries risks. The legal profession will need models in which AI handles routine work, lawyers intervene when judgment or representation becomes necessary and courts provide usable pathways for people who arrive with machine-generated assistance.

The profession can help design that future or spend its time defending barriers that technology and economics will eventually route around.

If it Can Happen to Lawyers

It would be easy to enjoy this story as an overdue reckoning for a profession known for high fees, slow change and layers of self-protection.

That would be a mistake.

Lawyers have more institutional protection than almost any other knowledge worker. They have licenses, bar associations, ethical rules, courts and laws prohibiting unlicensed competitors from practicing their profession. The government enforces the boundaries around their work.

If AI can separate legal services from lawyers despite all those protections, what profession is genuinely safe?

Accounting, consulting, financial analysis, software engineering, marketing, journalism, design, medicine and management all contain work that can be described, decomposed, evaluated and delivered digitally.

AI does not have to eliminate these professions to transform them. It only has to automate enough of their salaried or billable work that fewer people can produce substantially more.

That compresses fees. It reduces entry-level positions. It moves value away from production and toward supervision, judgment, trust and accountability. It leaves institutions employing fewer people even when the profession itself continues to exist.

The legal industry’s current struggle with hallucinated cases is therefore not merely a courtroom story. It is an early negotiation over who will control a profession’s intelligence once machines can produce much of its work.

Everyone wants the productivity. Law firms want AI to reduce research and drafting time. Clients want faster and less expensive answers. Courts want help managing overwhelming caseloads. People who cannot afford lawyers want meaningful access to the system.

No one wants to become the layer that gets commoditized.

Judges may be today’s AI firewalls, and lawyers may remain responsible for everything submitted in their names. But neither sanctions nor disclosure rules will settle the larger question.

What remains uniquely valuable about a lawyer when everyone can have a legal AI of their own?

And before the rest of us become too comfortable watching lawyers wrestle with that question, we should ask what remains uniquely valuable about the work we do.

If AI can unbundle a profession protected by licenses, courts, bar associations and the force of law, it can happen to yours.

And if it can, it will.