Rights and Bureaucracy · Blog · 🇧🇷 Brazil
Itaú puts AI in legal to read more than 90 million documents and run 2 million tasks a month
The legal solution, under the IARA platform with OpenAI models among others, reports efficiency gains of 20% to 80% in process stages.
- Profession
- Lawyer, Technology Manager
- Tools
- OpenAI, IARA, GPT
- Source
- openai.com ↗
- Country
- 🇧🇷 Brazil
- Report date
- October 16, 2026
- Verified on
- November 5, 2026
See more: AI for lawyers
What they did
In Itaú's legal department, an AI solution automates reading, analysis and interpretation of documents under IARA, the bank's internal governance and model-orchestration platform. OpenAI is among the providers available in that environment. The same OpenAI case describes ia.i, the bank's conversational layer that reached about 300,000 customers in the first phase, with a goal of 100% of the base by the end of 2026, and Customer DNA, financial context for about 70 million customers. The legal focus is volume and control: documents stay in the bank's context and model access goes through token cost, quality and responsible AI policies.
How it went
According to the case published by OpenAI, the legal solution runs on a repository of more than 90 million documents and executes more than 2 million tasks per month, with efficiency gains of 20% to 80% in stages of the department's processes. IARA concentrates observability, cost control and Responsible AI so the bank can run models in a regulated sector.
Good to know
This is an OpenAI case (also echoed by Itaú's own ICTi). The efficiency percentages cover specific stages, not the entire legal department. Expectations for manager productivity (more time with customers, faster service) are projections in the case, not proven results across the whole operation, as ICTi itself notes.
The prompt
I am a lawyer in [legal area] and need to review this set of documents [paste excerpts or metadata with no sensitive customer data]. Summarize what each one covers, point out clauses or passages that match [topic or risk], list inconsistencies among them and build a checklist of what a person on the team still needs to verify before any decision. Do not invent a fact that is not in the material and mark as uncertain anything that is incomplete.
Replace what is in [brackets]. Prompt written by us from the account, not a literal copy of the source. Do not paste other people’s personal data (privacy laws such as LGPD and GDPR apply).
Summary in our own words. Results as stated in the original account; we have not tested or audited them. Read the source ↗ Found a mistake?
Similar cases
01Habeas corpus written only with AI and a letter from the STJ to the OAB
An STJ justice sent a letter to the OAB over a habeas corpus petition with AI-fabricated citations and no original argument.
02Lawyer Fined by the TSE for Case Law Invented by AI
A TSE justice fined a lawyer and notified the OAB after nonexistent citations created with AI.
03When the "Rapporteur" Cited by AI Was a Bar Owner
A judge in Concórdia (SC) discovered that the rapporteur of a ruling cited in a filing was actually a bar owner.
04Case Law Invented by AI and the Fine of Ten Minimum Wages
A lawyer from Juazeiro (BA) was fined over a filing with fake rulings generated by AI and blamed the document on an intern.
05The ChatGPT-written petition that earned a fine from the TSE
A lawyer filed a petition written with ChatGPT at the TSE and was fined for bad-faith litigation.
06What happens when AI invents case law (TJSC)
A court fined a party who filed an appeal with nonexistent rulings and legal scholarship generated by AI.
07Porto Alegre tests an AI agent that checks public contracts in 2 hours instead of 35
In Procempa's OpenLab, the startup Inup built with the city's Comptroller General an agent that validates checklists for outsourced labor contracts.
08Disputing a denied flight delay compensation claim
The agent checked the flight data, concluded the airline's excuse didn't hold up under European rules and drafted the appeal.