BankingNewsAI Daily Brief ·
EU Commission issues AI Act guidance defining which systems qualify as high-risk.
Banking AI
Financial institutions & fintech technology
Citizens Bank picks MANTL to speed digital deposit account opening
Citizens Bank selected MANTL to modernize retail and business deposit account opening, targeting faster onboarding and improved digital conversion. The concrete move here is a named bank swapping in a specialized onboarding platform—one of the few areas where AI-assisted identity/KYC, decisioning, and workflow automation can move NIM-impacting volume quickly.
Action
Benchmark your own account-opening funnel (time-to-open, abandonment, manual review rate) against what a MANTL-style stack can deliver, then decide whether to buy, partner, or rebuild. Tighten model governance around KYC/IDV automation so faster onboarding doesn’t spike fraud losses or trigger remediation.
EU Commission publishes new AI Act guidance on what counts as “high-risk” AI systems
The European Commission released updated guidelines on classification of high-risk AI systems under the EU AI Act. For banks, this changes the practical boundary of which models and use-cases fall into the strictest compliance regime (documentation, risk management, human oversight, post-market monitoring), affecting both in-house builds and vendor tools.
Action
Map your AI inventory to the updated classification guidance and re-tier projects (credit, fraud, AML, collections, HR) into “high-risk vs not” with a named accountable owner per system. Insert AI Act classification and evidence requirements into vendor contracting now (right to audit, technical documentation, incident reporting).
General AI
Large language models & AI infrastructure
GitHub Copilot’s shift toward token-based billing is a real cost-risk signal for enterprise AI budgets
GitHub Copilot’s newer token-based billing is triggering backlash because it can make previously predictable developer-assistant spend variable and harder to cap. This mirrors a broader enterprise pattern: as teams move from “assistants” to agents and heavier model usage, unit economics can swing fast and break annual budgets.
Action
Put hard spend guardrails on AI dev tooling now: per-team token caps, policy-based model routing (cheap model by default), and chargeback so business units feel marginal cost. Negotiate pricing that aligns to outcomes (seat + capped usage tiers) before rolling agentic coding broadly across engineering and automation teams.
NVIDIA releases Cosmos 3 as an open “world model” platform—accelerating physical/vision AI beyond chat
NVIDIA launched Cosmos 3, an open model family aimed at physical AI reasoning, world simulation, and action prediction, with distribution via Hugging Face. Even if you’re not building robots, this signals the next wave of multimodal foundation models optimized for perception + planning, not just text—likely to spill into surveillance, branch security, and industrial ops ecosystems banks rely on.
Action
Assess where your bank is exposed to third-party “physical AI” (ATM fleets, cash logistics, branch cameras, facility access control) and ensure vendor roadmaps include auditable model behavior and incident response. Build internal capability to evaluate multimodal models (video/image) with the same rigor as LLMs, since risk will move off the keyboard and into the physical world.