BankingNewsAI Daily Brief · Friday, March 27, 2026
Banking AI
Financial institutions & fintech technology
Bank of America/Merrill productizes an end-to-end “AI meeting journey” for advisors with quantified time savings
Merrill and Bank of America Private Bank launched an AI-powered workflow that covers the full client-meeting lifecycle (prep through follow-up). The bank claims the tooling can save advisors up to four hours per meeting across millions of meetings annually, signaling a shift from generic copilot experiments to hard-ROI, process-embedded AI in front office.
Action
Quantify advisor-time economics and standardize an AI meeting workflow (agenda, research pack, note capture, follow-ups, CRM updates) with controls for suitability, recordkeeping, and supervision—this is becoming a competitive baseline for wealth productivity.
UK FCA starts using AI in supervision and authorizations as consumer AI-driven advice risk rises
The UK Financial Conduct Authority says it is accelerating internal AI use to speed up firm authorizations and to identify risks earlier, explicitly citing evidence that consumers are using AI tools to choose financial products. This is a regulator telegraphing both supervisory modernization and heightened scrutiny of AI-influenced advice/suitability outcomes.
Action
Stress-test your advice and product-journey controls for “customer used AI” scenarios (prompt evidence, disclosures, suitability checks, complaint handling) and assume faster, more data-driven FCA engagement on authorizations and supervisory requests.
General AI
Large language models & AI infrastructure
DeepMind ships Gemini 3.1 Flash Live: materially better real-time voice with lower latency
Google DeepMind released Gemini 3.1 Flash Live, targeting more natural, reliable audio interactions with improved precision and reduced latency. The practical change is that real-time voice agents become less brittle—fewer mis-hearings, tighter turn-taking, and more usable “live” customer conversations.
Action
Accelerate evaluation of voice-first agent use cases (contact center, collections, card disputes) and update your vendor due diligence to include latency, interruption handling, and transcript fidelity as first-class SLAs.
Google adds “switching tools” to move chat history/personal data into Gemini, raising the stakes in enterprise lock-in
Google is launching tools that let users transfer chats and personal information from other chatbots directly into Gemini. That lowers switching friction and signals the next battleground: proprietary conversation history and user context as the moat, not just raw model quality.
Action
Treat conversational data portability as a strategic control point—define retention, export, and migration policies now so you don’t get trapped in one assistant vendor’s memory layer.
OpenAI publishes its “Model Spec” as a public framework for model behavior and safety tradeoffs
OpenAI detailed its Model Spec—how it intends models to behave, how it balances safety with user freedom, and how it operationalizes refusals and boundary-setting. For enterprises, this is effectively a vendor-facing behavior contract that will influence auditability, policy alignment, and incident response when models do the wrong thing.
Action
Map your internal AI policy (acceptable use, refusal handling, escalation paths) to vendor behavior specs and require model-behavior documentation in procurement so legal/compliance aren’t reverse-engineering safety posture after deployment.