BankingNewsAI Daily Brief ·
The Financial Stability Board put agentic AI controls on the global supervisory agenda.
Banking AI
Financial institutions & fintech technology
FSB just put “agentic AI” controls on the global supervisory agenda (draft sound practices out for consultation)
The Financial Stability Board released draft “sound practices” for responsible AI adoption in financial services and opened a consultation. The draft explicitly calls out agentic AI and pushes firms toward stronger governance, model risk management, third‑party oversight, and operational resilience controls as autonomy increases. This is an early signal of what supervisors will expect to see evidenced in audits and exams across major jurisdictions.
Action
Stand up a cross-functional response (Risk, Compliance, Tech, Ops) to map the FSB practices to your current AI/Model Risk framework, with a specific gap assessment for agentic workflows (tool use, permissions, change control, kill-switches, logs). Use the consultation window to influence practicability (e.g., evidence standards, third-party model transparency) while aligning internal controls before expectations harden into supervisory checklists.
Scotiabank is formalizing “Scotia Intelligence” to scale enterprise AI—signal that Big Bank adoption is moving from pilots to operating model
Scotiabank announced expanded enterprise AI capabilities under “Scotia Intelligence,” positioning it as a bank-wide program rather than scattered use cases. The key change is organizational: central capabilities, governance, and enablement to accelerate adoption across functions. This is the playbook shift other large banks are making as AI becomes a productivity platform with controls, not an innovation lab experiment.
Action
Treat this as competitive confirmation that enterprise enablement (secure platforms, reusable components, governance-by-default) is now the differentiator, not isolated models. Benchmark your internal AI operating model (platform, data access, approvals, monitoring) against peers and accelerate standard “safe-to-deploy” patterns for agents, copilots, and analytics.
General AI
Large language models & AI infrastructure
OpenAI is acquiring Ona to turn Codex into long-running, enterprise-grade agents (persistent environments + governance)
OpenAI announced plans to acquire Ona to expand Codex with secure, persistent cloud environments that support long-running agents across enterprise workflows. That’s a step change from “chat + code suggestions” to agents that can keep state, run jobs over time, and operate inside controlled execution sandboxes. For enterprises, it reduces the gap between experimentation and production automation—while raising expectations for identity, audit logs, and permissions.
Action
Prepare for developer and ops demand to use persistent coding/automation agents by defining guardrails now: least-privilege credentials, environment isolation, mandatory logging, and change-management hooks. If you don’t offer an approved internal path, teams will route around controls with shadow tooling.
OpenAI models and Codex are now purchasable via Oracle Cloud commitments (enterprise buying motion just got easier)
OpenAI announced customers can access OpenAI models and Codex through Oracle Cloud, using existing Oracle commitments. This lowers procurement friction for large enterprises that are already standardized on Oracle commercial terms and governance. It also signals OpenAI’s continued push into “buy through your cloud provider” routes that CIOs prefer for controls, billing, and compliance.
Action
If you’re constrained by vendor onboarding and budgeting cycles, evaluate whether cloud-commit procurement (Oracle here, others likely to follow) can accelerate approved AI adoption without net-new contracts. Standardize a reference architecture for identity, network controls, and logging so business units can consume models through cloud marketplaces safely.