AI Regulation Tracker  /  Professional guidance

Sao Paulo's bar makes law-firm partners answerable for junior lawyers' AI use

Brazil's bar has set an ethics line on generative AI in legal practice. Partners and managers now carry a supervision duty over associate and staff AI use, and no automated output may go to a court without full human review.

By Anthony Guerriero  ·  Reviewed by The Leveraged Years Editorial Desk  ·  Published July 7, 2026  ·  Last updated July 7, 2026

The Ordem dos Advogados do Brasil has drawn an ethics line around how lawyers use generative AI, and in Sao Paulo it points straight at the people who run law firms. At the national level, the Conselho Federal da OAB (CFOAB) released the final version of its general recommendation on generative AI in legal practice, a set of principles that applies to every Brazilian lawyer. The harder duty is separate and local: an opinion from the OAB-SP (the bar's Sao Paulo section) Tribunal de Etica e Disciplina, approved on April 16, 2026, went further by putting a supervision duty on partners and managers and requiring human review of AI output before anything is filed. That partner-supervision mandate is the Sao Paulo tribunal's, not a national CFOAB rule.

The four pillars

According to the OAB, the CFOAB recommendation is organized in four parts: the legislation that applies to AI use, confidentiality and privacy, ethical legal practice, and communication about the use of generative AI. The document frames itself as guidance grounded in the profession's code of ethics rather than a penalty schedule. In the OAB's own account of the release, the rapporteur described alerts about the ethical duties of the profession as the central basis of the recommendation, and the coordinators stressed using the tools with responsibility and care.

That framing matters. The recommendation itself reads as a professional-conduct standard, not a statute. It tells lawyers what careful practice looks like and leaves enforcement to the ordinary disciplinary machinery of the bar.

Where the hard duty lives

The sharper obligation comes from the OAB-SP Tribunal de Etica e Disciplina. Its opinion, approved unanimously on April 16, 2026, sets two operational rules. First, partners and managers must supervise the AI use of the associates, trainees, and non-lawyer staff they oversee. Second, all AI output must be reviewed by a person before it reaches a court. The opinion treats reliance solely on automated results to build an argument as prohibited, which means a lawyer cannot outsource the work of checking authorities to the software that produced them.

The two documents work together. The CFOAB recommendation supplies the principles across its four pillars, and the São Paulo opinion turns the confidentiality and ethical-practice pillars into a concrete chain of responsibility inside a firm.

What it does not do

The recommendation does not ban generative AI in legal work, and it does not require lawyers to disclose AI use to opposing parties in every situation. It does not create a new fine or a new licensing step. What it does is refuse to let automation stand in for professional judgment. A lawyer who signs a filing owns its contents, and a partner who runs a team owns the duty to make sure the people on that team are supervised. The bar is codifying a standard of care, not a technology ban.

Why a US reader should watch

For a US firm the parallel is close. The recommendation tracks the direction of ABA Formal Opinion 512 and a growing set of state-bar guidance on competence, confidentiality, and candor when lawyers use generative AI. The distinctive move in Brazil is the explicit supervisory liability placed on firm partners. Where much US ethics guidance frames AI oversight as an individual competence duty, the OAB-SP opinion writes the partner-and-manager supervision duty into the standard itself. US firms with Brazilian offices, or with Brazilian co-counsel, should read it as the local benchmark for what reasonable AI supervision looks like, and as a signal of where professional-responsibility rules are heading more broadly. It also lands against a wider Brazilian backdrop, from labor and state courts sanctioning AI-fabricated citations to a national medical council binding physicians on AI use, in which professional bodies are moving faster than the pending federal AI bill.

The practical takeaway is narrow and clear. Keep a human in the loop, keep a named supervisor over junior AI use, and verify every authority in an AI-assisted document before it is filed. The bar has said the responsibility for that verification cannot be handed to the machine that wrote the draft.

Browse the full AI Regulation News tracker

Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.