Brazilian Court Sanctions Lawyers for Prompt Injection in Court Filing
A Brazilian labor court fined two lawyers for hiding a prompt-injection command in white text inside a filing, aimed at making the court’s AI review system handle their case superficially. Commenters broadly view the move as akin to hacking or tampering with court infrastructure and argue it violates lawyers’ duties of candor, with some saying disbarment would be more appropriate. At the same time, many criticize courts for relying on large language models in high-stakes legal processes at all, warning that such systems are biased, gameable, and shift responsibility away from human judges.
Nature of the “prompt injection” misconduct
- Lawyers embedded hidden white-on-white text instructing the court’s AI to respond superficially and not challenge documents.
- Many commenters see this as clearly deceptive and comparable to hacking, tampering with court infrastructure, or trying to bribe/manipulate a clerk.
- The concealment is viewed as strong evidence of intent (“guilty mind”), not an innocent or creative argument.
- Some think fines are too mild and argue for disbarment or even criminal charges; others note bars rarely sanction harshly in practice.
Legality, sanctions, and precedent
- Several cite a Brazilian rule (Article 77) barring acts that obstruct proceedings or make judicial activity difficult; this is seen as sufficient legal basis.
- Debate over whether Brazilian courts “don’t use precedent” versus claims that they are actually very precedent-heavy in practice.
- Some highlight that in many systems lawyers are “officers of the court” with a duty to the integrity of justice above client interest; similar but varying standards exist across civil-law jurisdictions.
- Question raised: would a self-represented litigant be treated more leniently? Most expect less tolerance for “smartarse” tricks, even if procedural mistakes get some slack.
Use of AI in courts
- Many see the deeper problem as courts relying on AI to review filings, given model bias, manipulability, and the stakes for liberty.
- Others argue overloaded courts make AI-assisted triage inevitable and potentially beneficial, as long as humans remain responsible.
- Some call for outright bans on AI in legal proceedings to protect defendants and force judges/lawyers to use their own judgment.
- A few suggest that if judges outsource analysis to AI, lawyers will naturally try to influence that AI, but others say influence must stay within proper legal advocacy, not hidden instructions.
Is “hacking the legal system” ever acceptable?
- One minority position: trying to “hack” the court’s AI should not be sanctionable; instead, courts should secure their systems.
- Majority response: sanctioning attempted hacks is itself part of securing systems; “trying to hack the legal system” is inherently improper.
- Comparisons made to SQL injection or buffer overflow exploits in court software: viewed as obviously punishable.
Technical mitigations
- Suggestions: render filings as images and run OCR / vision models to strip hidden metadata and invisible text, while noting this is not a complete defense.
- Some doubt the attack even worked, claiming modern LLMs are less susceptible to naive injections, though that’s not verified in the thread.