Brazilian court fines company for AI-generated legal petition without review
A Brazilian labour court fined a security and cleaning company for submitting an AI-generated appeal petition without human review, finding the system produced generic, inaccurate arguments that delayed proceedings and violated proper legal standards. The judge imposed fines totaling 7% of the case value for dilatory conduct and bad faith litigation.
What happened
A security and cleaning company submitted an appeal petition to the 3ª Vara do Trabalho de Mogi das Cruzes in São Paulo that was generated using artificial intelligence without critical review. The petition sought to contest supposed defects in a labour court judgment.
The AI-generated text was generic, standardized, and superficial, addressing issues not present in the original decision such as indirect dismissal and requesting compensation for items never included in the ruling, including remunerated weekly rest days. Judge Matheus de Lima Sampaio found that 'the AI did not read the process carefully, does not know specific legal concepts and is not capable of analyzing the peculiarities of the concrete case.'
Reported impact
The court found that the use of unreviewed AI-generated text improperly delayed the judicial proceedings by submitting documents lacking technical rigor and analytic depth.
Index analysis
The court distinguished between beneficial use of AI tools in legal work when used consciously by lawyers versus misuse that delays processes. Judge Sampaio stated that 'it is not permissible for the legal professional, whether or not using artificial intelligence, to submit to the Judiciary unreviewed texts that do not align with the concrete case, improperly occupying the time of the judge and the Judiciary with superficial expedients, devoid of the technical rigor and analytical depth that legal activity requires.'
Governance controls
The company was ordered to pay a fine of 2% of the updated value of the case for dilatory conduct and 5% for bad faith litigation, with amounts reversed to the opposing party.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.