A Brazilian bank just showed everyone how hard AI governance actually is
Itaú, one of Brazil’s largest banks, built two separate systems to run AI across its entire organization: a conversational interface called ia.i for customers and employees, and a governance platform called IARA to manage it all. They operate in one of the most regulated financial sectors in the world. That context matters.
If your reaction is “that’s a big bank, not my problem,” I’d push back on that. The decisions they had to make, who controls the AI, what it’s allowed to say, how you audit it, apply to any business deploying AI tools. The scale is different. The questions are the same.
Most small businesses skip the governance part entirely
This is where things get uncomfortable. A lot of small and mid-size businesses are already using generative AI. Staff are using ChatGPT to draft emails, write proposals, summarize documents. Sometimes with company data in the prompt. Almost never with a policy governing any of it.
Itaú built IARA specifically because deploying AI without a control layer is a liability in a regulated environment. Banking is obviously more regulated than most industries. But healthcare, legal, accounting, and financial services businesses in NJ face real compliance exposure when AI tools touch client data without oversight.
The practical question isn’t whether to use AI. It’s whether you’ve thought through what happens when it goes wrong, who’s responsible, and whether your current IT setup can even answer those questions.
What “AI First” actually requires from your IT foundation
Itaú’s strategy is called AI First, which sounds like a marketing phrase until you realize what it demands operationally. A conversational AI layer touching millions of customers requires identity controls, audit logging, access policies, and a way to update or retract AI behavior quickly. That’s not optional at their scale, and it’s not optional at yours either if you’re serious about using AI in production.
For a small business, the foundation looks like this:
- Controlled identities: who has access to which AI tools, and under what conditions
- Data boundaries: clear rules about what company or client data can enter an AI prompt
- Audit trail: some record of how AI is being used, especially for regulated data
- A way to change course fast if a tool behaves unexpectedly
None of this requires a platform the size of IARA. But it does require intentional setup, not just downloading an app and hoping for the best.
If you’ve read our post on how AI is moving faster than any government can plan, you already know that regulatory guidance is lagging well behind what tools can do. That gap puts the responsibility on businesses to set their own guardrails now.
The IT readiness problem most owners don’t see coming
AI tools generate and consume data. If your IT environment isn’t organized, that data goes somewhere you’re not tracking. We’ve also covered how AI-written code is changing what “ready” even means for a business IT setup. The same logic applies to AI-assisted operations. The tool is only as safe as the environment it runs in.
Identity management is a good place to start. If you’re using Microsoft 365, tools like Microsoft Entra ID and Intune give you real control over what devices and users can do. That’s the same control layer, at a much smaller scale, that Itaú is building with IARA.
Getting that foundation right before you expand AI use is a lot cheaper than cleaning up after a data incident. If you’re not sure where your gaps are, an IT assessment with Exine is a practical place to start.
FAQ
Do small businesses actually need an AI governance policy?
Yes, especially if staff are using AI tools with client or company data. A policy doesn’t have to be complex. It should define which tools are approved, what data can be used in prompts, and who is responsible when something goes wrong. Without it, you have no consistent standard to enforce.
What is an AI governance platform and do I need one?
An AI governance platform, like the IARA system Itaú built, manages oversight, access, and accountability for AI use across an organization. Small businesses don’t need enterprise-grade platforms. They do need clear policies, identity controls, and basic audit capability, often built on tools they already own like Microsoft 365.
How do I stop employees from putting sensitive data into ChatGPT?
Technical controls help more than policy alone. Microsoft 365 includes data loss prevention features that can flag or block certain content. Conditional access policies through Microsoft Entra ID can restrict which tools are accessible on company devices. A written policy backed by enforcement is more effective than a written policy alone.
Is generative AI safe to use in a regulated industry like healthcare or finance?
It can be, but the default settings on most consumer AI tools are not designed for compliance. You need to verify data processing agreements, confirm where data is stored, and ensure the tool fits within your regulatory framework. Using Microsoft 365 Copilot under an existing enterprise agreement is generally a safer starting point than unmanaged consumer tools.