Brazil's Itaú ships customer-facing generative AI as EU guidelines add agentic rules

Generative AI reaches the full customer base at Itaú

On 17 October 2026, Itaú outlined how it is applying generative AI to its entire customer base rather than limiting the technology to back-office pilots. The bank is layering a conversational interface called ia.i on top of an internal governance platform called IARA, framing the work as part of an AI First strategy that touches both customers and employees. For practitioners tracking enterprise rollouts, the concrete fact to file is that one of Brazil's largest banks has moved a generative-AI interface in front of retail customers, with a named governance layer sitting behind it. That shifts the deployment question from "can this be done?" to "what does the audit trail look like?".

EU research guidelines add an agentic-AI section

The European Commission's research and innovation directorate updated the ERA Living Guidelines on the responsible use of generative AI in research, refining them to address rapid advances in the field and now include guidance on agentic AI as well as more detailed recommendations. The update reads as a maintenance release for a framework its authors describe as "living": existing guidance has been edited in place rather than replaced, and the addition of an agentic-AI section suggests the maintainers now treat autonomous tool use as a routine research scenario rather than an edge case.

Practitioners running agents against internal data should re-read the guidance section that touches agentic AI specifically, because compliance language written for chat assistants does not automatically fit systems that act on the user's behalf.

Creative pipelines become an end-to-end product question

A 17 October 2026 live session hosted by a creative-AI lead at Superside pitched the audience on building a complete creative pipeline in a single workspace, covering campaign concepting and social creative. The framing matters because it treats generative tooling as a pipeline problem — concepting, production and distribution chained together — rather than a collection of one-off prompts. For practitioners, the implication is that prompt hygiene and asset hand-off now have to survive the boundary between brief, generation and export. Workflows that cannot pass clean metadata from one stage to the next will need to be patched.

Publishers experiment with AI for subscriber engagement

CNN and Google were reported on 17 October 2026 to be bringing generative AI into the audience engagement game, with the framing that the technology can help improve newsroom operations, retain subscribers, and reach new audiences. The detail worth carrying is the audience-engagement frame: a publisher working with a platform vendor is testing whether generative AI can intervene at the subscription boundary, not just the drafting boundary. That is a different surface from a model release or a governance update, and it changes which teams get pulled into procurement.

What to do next

Treat the Itaú launch and the EU guideline refresh as a paired signal: customer-facing generative AI is now a live production surface, and the maintainers of one European research framework have already extended their guidance to cover agentic systems. Practitioners should pull the latest ERA Living Guidelines section on agentic AI into their next governance review, and anyone shipping a customer-facing assistant should map their current audit trail against the IARA-style governance structure one large bank has now made public.

For practitioners building or buying agentic systems today, the open question is how their logging, consent and escalation steps translate to a guideline written before multi-step tools were routine — no public deadline has been set for the next revision. Keeping internal mock data and placeholder content production-ready remains a recurring need; teams should make sure pipelines can generate realistic, non-sensitive test corpora at scale rather than scraping production logs.