AI Regulation Tracker / Enforcement action
Brazil's Paraiba Bar Suspends Two Lawyers Over Hidden AI Prompt-Injection Commands in Court Filings
On Tuesday, July 14, 2026, the OAB-PB ordered a precautionary suspension of two lawyers accused of embedding hidden instructions in court filings to manipulate the judiciary's AI drafting tools. It is an interim disciplinary measure, not a final judgment, and it looks like the among the first documented Bar suspensions aimed squarely at prompt injection.
On Tuesday, July 14, 2026, the presidency of the OAB-PB, the Paraiba state chapter of the Ordem dos Advogados do Brasil, ordered a precautionary suspension of two lawyers from the practice of law and opened an ethics-disciplinary process against them. The stated reason is that they allegedly planted hidden commands inside court filings to manipulate the judiciary's artificial intelligence. The legal footing is the Estatuto da Advocacia, Lei 8.906/94, the statute that governs how lawyers in Brazil are licensed and disciplined.
What the lawyers are accused of doing
The technique at the center of this is prompt injection. In plain terms, that means slipping instructions into a document that are meant to be read by a machine and acted on, while staying invisible or nearly invisible to the human reading the same page. According to the OAB-PB, the two lawyers did this with the classic toolkit: white text on a white background, type shrunk to a microscopic size, and words deliberately broken into fragments so a casual reader, or a plagiarism check, would not catch them, but a language model parsing the raw text would.
What the hidden text reportedly said is the part that makes this more than a curiosity. The embedded instructions told the AI to ignore impartiality and directed that the appeal be granted. One filing even included a note describing itself as a test to see whether the judge was relying only on AI to decide. The target was the state court's generative-AI drafting system, the tool judges in Paraiba use to help produce draft decisions. The alleged goal was to get the machine to write a favorable draft the judge might then adopt.
That is a different animal from the AI cases that have made headlines so far. Most of those involved lawyers who used a chatbot to write a brief, the chatbot invented case citations that did not exist, and the lawyer filed them without checking. Those are failures of diligence. This is being framed as something more deliberate: an attempt to reach into the court's own tooling and turn it against the other side. The OAB-PB is treating it as an attack on the integrity of the process, not as sloppy research.
Where the cases came from
The two matters reached the Bar by different routes. One was flagged by a judge working in the Nucleo de Justica 4.0, the branch of the Paraiba court system built around digital and AI-assisted case handling, who noticed the hidden commands and referred the conduct. The other traces to a case in Sousa, in the interior of Paraiba, where a judge levied fines totaling R$32,800 and referred the lawyer not only to the OAB but also to the Policia Federal and the Ministerio Publico Federal, the federal police and federal prosecutors. In other words, the same conduct is being examined through more than one lens at once: professional discipline through the Bar, and potential criminal or civil exposure through the federal channels.
I want to be careful about what is and is not public here. The disciplinary process is running under judicial secrecy, segredo de Justica, and the OAB-PB has not published the lawyers' names or the process number. So I am not naming them, and neither should anyone reporting on this responsibly. It is also worth noting that one of the lawyers has publicly denied acting with intent to manipulate the court. That denial is part of the record, and it is exactly why a precautionary suspension is not a verdict.
What this is, and what it is not
This matters, so read it slowly. A precautionary suspension is an emergency measure. It stops the lawyer from practicing while the disciplinary process runs, because the Bar has decided the alleged conduct is serious enough that letting the person keep practicing in the meantime is a risk. It is not a finding of guilt, it is not a disbarment, and it does not carry a fixed term. The OAB-PB itself put it plainly in its statement: the precautionary suspension does not represent a definitive judgment, but an exceptional measure provided for in the Estatuto da Advocacia.
The president of the OAB-PB, Harrison Targino, framed the order around the Bar's institutional duty to act firmly whenever there are signs of conduct that could compromise professional ethics, good faith in the process, and society's trust in the legal profession. That is the language of a regulator protecting the integrity of the system, and it tells you how the Bar is reading this: not as a technology story, but as a good-faith story that happens to involve technology.
So the honest summary is that two lawyers have been suspended pending a process, over serious allegations that they have not yet had a full chance to answer, one of them says he did not intend to manipulate anything, and the outcome is not decided. What is already established is the enforcement posture. The Bar has signaled that using hidden prompts to hijack a court's AI is the kind of thing that gets you pulled off the field while the referee sorts it out.
Why a US professional should care
The OAB-PB is not a US regulator, and nothing here binds a US practitioner or a US filing. The reason to pay attention is the theory, because the theory travels well.
US courts are adopting AI drafting tools on the same curve Brazil is. Once a tribunal, or an opposing counsel, is running a language model over the documents you file, a hidden instruction inside your document becomes a live attack surface. Prompt injection is already showing up in the wild in e-filing and discovery contexts, where one side wonders whether the other has salted a production or a brief with text meant to steer an automated reviewer. What Paraiba just did is put a name on the consequence: adversarial manipulation of a tribunal's AI through concealed prompts can be treated as bad-faith conduct serious enough to warrant an emergency suspension, before any final ruling.
For finance leaders and general counsel, the practical read is not about Brazilian bar rules. It is about controls. If your organization files documents into systems that run AI over them, or runs AI over documents other people send you, you now have a concrete example of the hidden-instruction risk being treated as sanctionable conduct rather than a clever trick. That is worth a conversation with the people who own your e-filing, your discovery review, and your document intake, before it is your problem.
What to do now
Read the finding for what it is: a precautionary suspension pending a disciplinary process, not a disbarment and not a final judgment. If your teams use AI to review filings, contracts, or discovery, assume documents can carry hidden instructions and check the raw text, not just the rendered page, for white-on-white text, microscopic fonts, and fragmented words. Keep a human accountable for any decision an AI tool drafts, so a manipulated draft cannot become a final action without someone signing off. And if you handle inbound documents, treat prompt injection as a known threat in your intake controls the same way you already treat malware in attachments. None of this is a legal requirement flowing from Brazil. It is a prompt to look at your own controls before the theory shows up in a court you actually appear in.
Questions professionals are asking
Were the two lawyers disbarred?
No. They received a precautionary suspension, an interim measure that stops them from practicing while a disciplinary process runs. It is not a disbarment, it carries no fixed term, and it is not a final judgment. The OAB-PB itself states the precautionary suspension does not represent a definitive ruling.
What is prompt injection in this context?
It is the practice of hiding instructions inside a document that a language model will read and act on, while keeping them invisible to a human. Here the lawyers allegedly used white text, microscopic fonts, and fragmented words to insert commands telling the court's AI to ignore impartiality and grant the appeal, aimed at the Paraiba court's generative-AI drafting system.
How is this different from the fake-citation AI cases?
The fabricated-citation cases involve lawyers who trusted a chatbot that invented sources, a diligence failure. This case is framed as a deliberate attempt to manipulate the court's own AI tooling to steer the outcome, which the Bar treats as a good-faith and integrity problem, not sloppy research. It appears to be the among the first documented Bar disciplinary suspensions aimed specifically at prompt injection.
Does this affect US lawyers or filings?
Not directly. The OAB-PB is a Brazilian state bar and has no authority over US filings. The reason to watch it is the transferable enforcement theory: as US courts adopt AI drafting tools, manipulating a tribunal's or an opponent's AI through hidden prompts could be treated as sanctionable bad-faith conduct.
Why are the lawyers not named?
The disciplinary process runs under judicial secrecy, segredo de Justica, and the OAB-PB has not disclosed the lawyers' names or the process number. Responsible coverage does not publish them. One of the lawyers has publicly denied intent to manipulate the court, which is part of why the suspension is precautionary rather than final.
RELATED BRIEFINGS
Browse the full AI Regulation News tracker
Informational analysis for working professionals, not legal advice. This describes an interim disciplinary action, not a final judgment, and the allegations have not been fully adjudicated. Confirm how any rule or professional duty applies to your situation with qualified professionals in the relevant jurisdiction.