What Artificial Intelligence can already do today for a labor-law firm, and what it still can't

In May 2026, Justice Rogerio Schietti Cruz of Brazil's Superior Court of Justice (STJ) denied a preliminary injunction in a habeas corpus case after finding that the petition cited 16 precedents whose reporting justice, deciding panel and content simply did not exist: entire passages "hallucinated" by a generative Artificial Intelligence tool. The justice ordered that the matter be formally reported to the Brazilian Bar Association (OAB) so it could take appropriate action, and summed up the problem in a line that should be posted on the wall of every firm already using AI: "technology serves the professional, but does not replace them or excuse them from checking what they sign."

This case isn't a reason to avoid AI in legal practice: it's proof that the risk lies in using it without criteria, not in using it at all. The OAB itself had already anticipated this: in November 2024, its Federal Council published formal recommendations on the use of AI in legal practice, covering applicable law, confidentiality, ethical practice of the profession and the duty to disclose when content was produced with the support of generative AI. In other words, the institution that regulates the profession already recognizes the tool as legitimate, as long as it is used under declared human supervision.

That said, what actually works today, in practice, at a labor and social-security law firm?

Faster case-law research, never without verification. Generative AI considerably speeds up the search for and initial summary of rulings, but the STJ case above shows exactly why every citation needs to be checked against the original source before it goes into a filing. The real gain is in first-read time, not in eliminating the check.

Organizing and screening large volumes. Sorting cases by legal theory, urgency or likelihood of settlement, and quickly flagging what needs immediate attention within a large caseload, is a task where AI already delivers concrete value: without making the decision, just organizing what used to take hours of manual sorting.

First pass through long documents. Pulling the relevant points out of a lengthy contract, a collective bargaining agreement or a bulky case file before the responsible attorney's own review is another already-mature use: it shortens the time to a first qualified read, it does not replace the qualified read itself.

Ongoing monitoring cross-checked against the caseload. Instead of manually tracking everything that changes at the TST, STJ and STF (Brazil's labor, superior and supreme courts), it is possible to cross-check those changes against the cases the firm already has underway: knowing in advance which real cases might be affected by a new ruling.

What these four fronts have in common: all of them speed up one stage of the work, and none of them signs the filing, decides the strategy or removes the need for review by whoever has the training and the professional responsibility to do it. It is that line, clear, and now also recognized by the OAB and reinforced by the STJ, that separates a real productivity gain from disciplinary risk.

Can your firm point today to exactly where AI is being used, and who checks what it produces before it gets signed?