Armundia Group CIO: How Banks Can Secure Customer Data While Using External LLMs

According to a post shared on LinkedIn by Armundia Group staff, Alessandro De Leonardis, CIO of Armundia Group, presented the paper "Hybrid Symbolic-Generative RAG for Know-How Protection and Privacy Preservation" together with Dario Iannella[2] at FAIEMA 2026 - the 4th International Conference on Frontiers of Artificial Intelligence, Ethics and Multidisciplinary Applications.[2]The post says the paper's subject was "using generative AI in financial services without exposing what is sensitive, client data, confidential documents, proprietary know-how."[2]According to the same post, an anonymisation layer replaces sensitive entities with neutral tokens before the call to the external LLM[2], and the real values are put back together only inside the bank's "trust perimeter."The post describes the proposal as "an architecture, not a model"[2] and says security "is the engineering around the model, designed to fail closed."[2]In an earlier LinkedIn post announcing the talk, De Leonardis wrote that the approach was "validated on a real-world banking use case and built on a Retrieval-Augmented Generation stack."[3]De Leonardis also wrote that in regulated sectors, data exposure "becomes the real dividing line between a project that can go into production and one that stays a prototype."[3]Separately, The Fintech Times reported that De Leonardis has argued automation in insurance brokerage stops where human judgement must explain, arbitrate and govern.[4]