The sequence matters more than either incident alone. A first-instance labor judge in Parauapebas identified an adversarial hidden prompt inside a petition and sanctioned the lawyers who filed it. Ten days later, a minister of the Superior Tribunal de Justiça formally notified the OAB after a habeas corpus arrived containing sixteen fabricated precedents generated entirely by AI without human review. Two judicial levels, two distinct categories of AI misuse, ten days apart — and no Brazilian statute yet defines either as a specific legal offense.

This is how legal norms form in civil law systems under pressure: courts act on principle, publish reasoned decisions, and those decisions become the substrate on which regulation is eventually built. The Brazilian law reacted before it could define the conduct, and what the Parauapebas and STJ rulings do is establish the enforcement template before the template has a statutory name. Any Brazilian legislator drafting AI-in-courts rules now is writing to ratify what judges already decided, not to set new standards.

The legal and professional consequences in both cases arrived at the same destination through different routes. Submitting a petition with hallucinated precedents is a failure of professional diligence — the attorney delegated a task and did not verify the output. Embedding an invisible adversarial prompt designed to manipulate a court's AI tool is something harder to characterize as mere negligence. It requires knowing that the court uses AI to process documents, understanding how that AI can be influenced through injected instructions, and deliberately encoding that instruction in a way invisible to human reviewers.

The Parauapebas judge treated the hidden prompt as an integrity violation rather than a competence failure, and that distinction carries forward into every subsequent Brazilian ruling on similar conduct. The Adrian Lerer analysis of the Parauapebas ruling frames it as a redefinition of what procedural loyalty means when AI mediates the process — once a court's AI infrastructure becomes a surface that attorneys can manipulate, the duty of candor expands to cover that surface. The framing does not require a new statute to be operative; it applies through existing rules of professional conduct.

The STJ minister's notification to the OAB is the institutional move that individual sanctions cannot replicate. When a court sanctions a single attorney, it creates a record and a precedent. When a supreme court minister formally contacts the profession's governing body, it is signaling that the profession itself has a compliance obligation — and that the courts will not indefinitely absorb the cost of AI misuse that the bar has declined to address proactively.

The OAB now has a formal record of a demand from the STJ. Whatever it does next — issue guidance, launch an inquiry, draft competency standards — that action will be traceable to a specific judicial trigger. The pattern of fabricated AI legal filings that U.S. federal courts have been managing through individual sanctions alone has not produced a professional standard. Brazil's courts appear to have concluded that individual deterrence is insufficient, and they escalated accordingly.

The broader conversation in Portuguese-language legal practice about AI limitations is not a translation of the English-language hallucination debate — it has a distinct technical character. Brazilian federal court citations follow specific formatting conventions, procedural histories, and naming structures that differ from the sources on which most general-purpose AI models were trained. An AI producing plausible-sounding STJ precedents is not generating random noise; it is generating artifacts that are specifically calibrated to appear credible within a legal tradition the model does not fully represent.

This specificity matters for how the OAB and Brazilian courts frame the professional standard going forward. The risk is not that attorneys are using AI — it is that they are using AI tools built primarily around other legal traditions and trusting the output without verification against Brazilian sources. The single word that machine translation misses in cross-jurisdictional legal documents is the micro-level version of the same problem: language that sounds correct can embed legal assumptions that do not transfer. In the STJ case, sixteen fabricated citations that sounded correct transferred assumptions that had no basis in Brazilian law.

The two Brazilian rulings function as regional precedent in a way that U.S. sanctions cases have not. The legal systems of Argentina, Colombia, Chile, and Peru share enough structural similarities with Brazil's civil law tradition that bar associations in those countries are watching the OAB notification with direct professional interest — not as observers of a foreign jurisdiction but as potential adopters of a framework their own courts have not yet been forced to develop.

The enforcement template is already in motion. Attorneys who file AI-generated documents without verification face sanctions. Attorneys who embed adversarial prompts face integrity charges, not negligence charges. Bar associations that have not issued guidance now have a formal STJ notification as the standard against which their inaction will be measured. Brazil's courts did not wait for legislators, and the rest of Latin America's legal profession will not get the luxury of watching legislators move first either — the judicial precedent is already there to be cited.