BankingNewsAI Daily Brief · Wednesday, March 25, 2026
Banking AI
Financial institutions & fintech technology
HSBC creates a Chief AI Officer role—signal that AI is moving from ‘tech program’ to operating model
HSBC appointed its first Chief AI Officer, elevating AI leadership beyond ad hoc ownership in technology or digital teams. The move formalizes accountability for AI strategy, governance, and execution at group scale.
Action
Establish (or revalidate) a single accountable AI exec owner with budget authority and measurable delivery KPIs, not just policy oversight. Use the CAIO construct to force consolidation of duplicate AI efforts across businesses, vendors, and model stacks.
Fraud/AML vendors are pivoting to tabular “foundation models”: Feedzai launches RiskFM for risk decisioning
Feedzai launched RiskFM, positioning it as the first tabular foundation model purpose-built for financial data and risk decisioning across fraud and AML. This is a meaningful shift from rules + classical ML toward reusable pretrained models designed for high-volume, structured transaction data.
Action
Pressure-test your fraud/AML roadmap: require vendors (and internal teams) to show measurable lift vs. current ML on your data, plus explainability and drift controls suitable for model risk governance. If you’re considering “agentic” workflows in financial crime ops, this is the kind of base model layer that will determine whether automation is safe at scale.
General AI
Large language models & AI infrastructure
Anthropic is expanding ‘computer use’ automation—Claude can operate apps and spreadsheets on a user’s machine
Anthropic says Claude can now use customers’ computers to complete tasks—opening apps, navigating the browser, and editing spreadsheets like a human operator. This increases real automation potential but also expands the security and control surface (credentials, approvals, logging, and data exfiltration paths).
Action
Accelerate your agent control-plane requirements: mandate step-level logging, permissioning, and kill-switches for any tool that can act on endpoints or privileged business apps. If you’re rolling out “digital workers,” align IAM and DLP to agent actions (not just human users) before broad deployment.
Databricks is consolidating AI security via acquisitions—signal that ‘security for AI’ is becoming a platform layer
Databricks acquired two startups (Antimatter and SiftD.ai) to underpin a new AI security product. The move suggests AI security is shifting from point tools to integrated controls embedded in the data/ML platform where models are built and deployed.
Action
Decide whether AI security lives primarily in your data platform (Databricks/Snowflake/etc.) or in standalone security tooling, then standardize—fragmentation will slow audits and incident response. Require your platform vendor to cover model/data access controls, monitoring, and policy enforcement end-to-end to avoid bespoke glue code.