AI is helping judges to quickly close cases, and lawyers to quickly open them (restofworld.org)

Brazil’s courts have embarked on one of the world’s largest deployments of AI to cope with an enormous caseload — roughly 76 million lawsuits nationwide — and it’s reshaping both adjudication and litigation. Since 2019 the judiciary has rolled out 140+ machine‑learning and LLM-based projects that search precedents, categorize matters, draft documents, predict rulings and flag repeat litigants. Tools like the Supreme Court’s MarIA (which leverages Google’s Gemini and OpenAI’s ChatGPT) and commercial platforms such as Harvey (built on OpenAI models fine-tuned on legal data) have helped judges and clerks process files far faster: nationwide case closures were 75% higher last year than in 2020 and the Supreme Court backlog fell to its lowest level since 1992. At the same time, more than half of Brazilian lawyers now use generative AI daily and filed 39+ million new lawsuits last year (a 46% rise since 2020), creating what officials call a “vicious circle” of faster closures and faster refilling of dockets. The consequences are technical, economic and ethical. AI improves efficiency and lowers drafting time from minutes to seconds, attracting >$1B in VC this year and a projected $47B legal‑tech market by 2029. But LLMs hallucinate: researchers estimate 350+ global cases of made‑up precedents (six in Brazil), prompting fines and UN warnings against “techno‑solutionism.” The practical takeaway for practitioners and policymakers: AI can automate many routine legal tasks, but outputs must be human‑reviewed, sources cross‑checked, and regulatory safeguards instituted to prevent errors and preserve justice’s contextual and normative judgments.