The Equalizer the Legal Profession Didn’t Ask For

AI is closing part of a justice gap the legal profession spent a century building. The record shows what its resistance is actually protecting.

The Gap AI Walked Into

I want to start with the number the legal profession does not lead with when this subject comes up. The Legal Services Corporation’s 2022 Justice Gap study found that low income Americans get no help or not enough help for 92 percent of the civil legal problems that substantially affect their lives. Nearly three quarters of low income households had at least one such problem in the prior year. Cost was the reason most often given for not even trying to find a lawyer.

Family court reflects the same math from the other direction. Research compiled by the Institute for the Advancement of the American Legal System puts the share of family cases with at least one unrepresented party at 80 to 90 percent in many courts nationally. Dockets built for two represented parties in front of one judge now routinely run with one side, or both, appearing alone.

This is the system generative AI walked into. A structure that assumes a lawyer on both sides of the table, serving a public where that assumption is now the exception rather than the rule. A tool that can draft a motion, explain a court form, or flag a missed deadline does not need to be perfect to change an outcome for someone who was otherwise filing nothing at all.

What the Record Actually Shows

The legal profession’s first big public reckoning with generative AI came in June 2023, when a federal judge in Manhattan sanctioned two attorneys five thousand dollars each. Their brief in Mata v. Avianca cited six court decisions. ChatGPT had invented all six, complete with fabricated quotes and case numbers. The episode became shorthand for why AI has no place near legal work.

Three years later, the record complicates that shorthand considerably. A running public database maintained by legal researcher Damien Charlotin now tracks more than two thousand court decisions worldwide where a party relied on AI generated hallucinated material, most of them in the United States. Self-represented litigants make up the largest share, but practicing lawyers account for hundreds of entries of their own, including attorneys representing some of the largest institutions in the country. One entry logged this month involves a law firm’s opposition brief against Brigham Young University, produced with four different AI tools and containing fabricated and misquoted case law. The sanction was three thousand dollars and two ethics courses.

The Bar Has Done This Before

None of this is the legal profession’s first fight over who gets to help people navigate a court without paying for a law degree to do it. In 1972, Michigan adopted no fault divorce, and at least two Detroit area businesses began selling people the forms and instructions to file their own. The organized bar moved fast. The State Bar of Michigan brought unauthorized practice proceedings against one firm. A court enjoined the other on its own initiative. Legal scholar Arthur Miller, writing in the University of Michigan Journal of Law Reform the following year, concluded the firms would likely lose under existing law, and questioned openly whether that outcome served the public or only the bar.

Fifty years later, the target moved from divorce kits to chatbots, and the reasoning did not move at all. In 2023, DoNotPay founder Joshua Browder scrapped a plan to have his company’s AI argue a traffic ticket case in court after, he said, a state bar official warned him that unauthorized practice of law carries up to six months in jail in some states. He later said the threat of jail time was, in his words, “enough to give it up.” California’s bar confirmed it had opened an investigation. Separately, an Illinois law firm sued DoNotPay on behalf of every law firm in the country, alleging its marketing amounted to false advertising. A federal court dismissed that suit for lack of standing. A different customer’s unauthorized practice and consumer fraud claims against the company settled in 2024.

Michigan’s own bar has not gone that far with generative AI, at least not yet. Its 2025 AI Workgroup report, a document its authors call a living work in progress, acknowledges that AI assisted legal help raises unauthorized practice questions under Michigan’s existing statute, and recommends more member education rather than more enforcement. Whether that restraint holds is worth watching against the state’s own history.

Same Mistake, Different Consequence

The clearest evidence that this is not really about accuracy is what happens after the mistake gets caught. When attorneys at established firms are caught filing AI hallucinated citations, the documented consequence is typically a monetary sanction in the low thousands and an order to complete continuing education. When self-represented litigants make the identical error, the outcome is often a warning, but sometimes something sharper: a referral to the state bar to investigate whether the person, the one without a law license, engaged in unauthorized practice of law on their own behalf. A California appellate court did exactly that this year, sanctioning pro se appellants for a brief full of fabricated citations and forwarding the matter to the state bar to investigate unauthorized practice.

New disclosure rules are extending that exposure further. Florida’s Eleventh and Seventeenth Judicial Circuits now require both attorneys and self-represented litigants to disclose AI use and certify its accuracy, with sanctions ranging up to contempt and disciplinary referral. That is a defensible rule for a licensed attorney, who carries a bar card, malpractice coverage, and a duty of candor to backstop the certification. Applied to someone with no lawyer, no law degree, and often no way to know what a fabricated citation looks like, it is a certification requirement with nothing underneath it.

The Institutional Response, Graded

Laid side by side, the record does not describe a system failing uniformly to catch a new kind of mistake. It describes a system that is quite good at one thing and largely indifferent to another.

The Legitimate Concern Underneath the Fear

None of this means the competence concerns are manufactured. A hallucinated citation in a real client’s brief is not a victimless error. It can cost a case, a deadline, or in immigration and criminal matters, someone’s liberty. ABA Formal Opinion 512 is a sound document. It tells lawyers that using generative AI does not suspend their existing duties of competence, confidentiality, and candor to the court, and that verifying AI output is simply part of the job now. Judges who require certification that a human reviewed AI generated content, starting with Judge Starr’s early standing order, are responding to a documented failure mode, not an imagined one.

The honest version of this argument is not that the profession’s concerns are fake. It is that the concerns are real and the enforcement energy behind them is not distributed according to those concerns. If accuracy were the only thing being protected, a large firm’s AI generated fabrications and a self-represented tenant’s AI generated fabrications would draw comparably serious institutional attention. The record so far says they do not.

Legal Knowledge Should Not Be a Class Privilege

There is something classist about resenting people for gaining access to knowledge they need to protect themselves. When an attorney’s objection is that a person without a law degree can now understand an argument, ask a better question, or challenge a claim they once would have accepted, I would call that professional sour grapes. Someone else’s growing ability to defend their interests should not be an injury to your professional identity.

No reasonable public education system could teach every citizen every procedural rule, filing requirement, and legal doctrine they might someday need. A person cannot be expected to leave high school prepared for every possible eviction, custody dispute, debt collection lawsuit, or fight over public records. People need ways to keep learning when those problems become real. Their ability to understand the rules governing their lives should not depend on whether they can afford someone else’s hourly rate.

A law degree represents substantial training. It deserves respect. So does the effort of someone learning enough to protect their home, their children, or their livelihood. AI does not automatically give that person an equal footing with experienced counsel. But getting closer to a fair chance is a legitimate aspiration, and attorneys should welcome tools that help people ask informed questions and recognize when they need further help.

Criticize fabricated citations. Demand accurate work. Explain where a tool falls short. But if the resentment begins simply because someone who could not afford your services has found another way to learn, the issue is class. The public should not have to remain confused to preserve the value of a professional credential.

What This Looks Like From the Self-Represented Side

I want to be direct about something the profession rarely says out loud. I have spent years doing legal and investigative work without a law license standing behind me: filing my own motions, requesting my own records, building timelines from documents nobody handed me. The tools that help with that work, AI included, do not replace judgment. They replace the first draft, the first search, the first read of an unfamiliar statute, the parts of legal work that used to require paying someone by the hour just to get started. That is not a small thing to take away from a fee structure built entirely around the hour.

None of this makes AI safe to use blind. Every fabricated citation in that database, whether it belongs to a self-represented person or a licensed attorney, is a failure of the same discipline: verify before you file, whoever wrote the first draft. But that discipline is the same discipline the profession is now telling its own members to learn for the first time. The rest of us have been practicing it out of necessity since before ChatGPT had a name.