New Banking AI Response Modes Give Banks Topic-Level Risk Control
Banks and credit unions can now vary how much freedom their AI gives an account holder depending on the question being asked. On 21 July 2026, Glia introduced three configurable banking AI response modes for Glia Banker, its voice and digital AI agent, letting an institution set how flexible or how tightly scripted a reply is on a topic-by-topic basis. The move reframes AI risk as something to be tuned rather than accepted or rejected wholesale, and it targets the single obstacle that has kept most banking AI stuck in pilot: the fear that a generative system will invent an answer to a question that carries regulatory or financial consequences.
What did Glia actually launch?
Glia added three interaction styles to Glia Banker, which the company says is used by more than 700 financial institutions. Each mode governs how much latitude the AI has when it responds.
Strict Mode is the existing behaviour those institutions already run: the agent returns answers exactly as approved inside the institution's own knowledge base, with no rewording. Rephrase Mode keeps the approved answer intact but adapts the phrasing to match the customer's language and context, so the reply reads more naturally without introducing any new information. Compose Mode goes furthest, generating replies in real time from institution-designated sources such as policy documents, product information and approved knowledge articles, while staying confined to that approved material rather than drawing on the open internet or a general-purpose large language model.
The practical design point is that an institution can mix the three across different topics. The same bank might run loan and wire-transfer inquiries in Strict Mode for full compliance control while allowing Compose Mode to handle low-stakes conversational topics such as travel notifications or branch queries. Glia says the modes are intended to handle up to 80% of inquiries so that human staff can concentrate on relationship-heavy work.
Why does topic-level control matter for banks?
The reason this lands as news rather than a feature note is that it addresses the specific failure pattern in enterprise generative AI. Across industries, the large majority of generative AI pilots never reach production, a problem Glia's own 2026 Banking AI Benchmarks Report puts at 95% and one that independent commentary through 2025 and 2026 has repeatedly attributed to governance and trust gaps rather than raw model quality. In regulated banking, the blocker is rarely capability. It is the inability to prove that an automated answer was correct, sourced and compliant.
By letting risk appetite vary by topic, Glia is betting that institutions will deploy AI faster when they no longer have to choose between a rigid script for everything and an open-ended model they cannot fully control. That is a meaningfully different proposition from an all-or-nothing rollout, and it maps to how banks already think about risk internally, where a balance transfer and a location lookup sit in entirely different tiers.
Where does this sit in Glia's wider position?
Glia is not a new entrant. Founded in New York in 2011 as SaleMove, the company rebranded to Glia and has raised roughly $152 million across seven rounds, most recently a $45 million Series D led by Insight Partners in March 2022 that pushed its valuation past $1 billion. Independent estimates put its annual recurring revenue near $90.8 million in 2024, up from about $73.1 million a year earlier. Its 2022 acquisition of banking chatbot specialist Finn AI gave it a purpose-built foundation in financial-services conversation, which is central to how it now differentiates from horizontal contact-centre and customer-experience vendors.
The company's stated commercial thesis is that specialised banking AI is what lets community and regional institutions compete with far larger banks and with fintechs, which by Glia's figures are capturing a large share of new checking accounts. The response modes extend that pitch by giving those institutions granular control over where automation is aggressive and where it is locked down.
What should institutions check before relying on it?
Two claims warrant scrutiny before a buying decision. Glia markets a contractual guarantee against AI hallucination and prompt-injection attacks, an unusual commitment in a market where most vendors offer best-effort assurances. The strength of that guarantee depends on its contractual scope and remedies, which are not public and should be examined directly. Separately, Compose Mode still generates language in real time. It is constrained to approved source material, but institutions will want to test how it behaves at the edges of that material before routing sensitive topics through it.
Why This Matters to FinanceX Readers
For finance professionals and technology buyers, the signal here is that the banking AI conversation is shifting from whether to deploy generative systems to how precisely their behaviour can be constrained. Vendors are increasingly competing on control and auditability rather than on raw fluency, and topic-level configuration is becoming a differentiator.
For community and regional institutions weighing AI against tighter budgets and heavier compliance loads, an adoption model that lets them contain risk exactly where it matters, rather than gate the whole programme, lowers the barrier to putting AI into production. For investors, it reinforces that the defensible position in financial AI is being built around governance, not just capability.
