BankingNewsAI Daily Brief ·
OpenAI introduces GPT-6.1 Sol
Banking AI
Financial institutions & fintech technology
Waller discusses AI agents in payments
For the payments head at a regional bank
Agent-delegated payments turn customer authority into a product-control and fraud-loss problem. Your bank needs to decide what permissions, funding limits and authentication it will require before agents can initiate payments.
Federal Reserve Governor Christopher J. Waller discussed AI agents in payments at Sibos 2026. He said agents can plan and execute multistep processes using large language models and transact autonomously. Waller described agent-assisted commerce, where buyers search with an agent but make payment decisions themselves, and agent-delegated commerce, where buyers authorize an agent to shop and pay within constraints. He gave the example of an agent using a pre-funded virtual card to buy groceries based on past purchases. Waller said consumer-to-business use cases will come first, with assisted models preceding delegated models because delegated purchases need more trust mechanisms and guardrails.
→ Action
Payments product: Define customer permissions, funding limits and authentication controls for agent-initiated card payments.
Read article → from Federal Reserve
Bowman says AI adds complexity to cyber risk
For the CIO of a community bank
AI turns phishing, vulnerability discovery and live attacks into faster threats. Put those scenarios into tabletop tests now, then test whether identity controls and patching can keep up.
Federal Reserve Vice Chair for Supervision Michelle Bowman told community banks that AI is adding complexity to cyber risk. She said threat actors can use AI to accelerate vulnerability identification, create sophisticated social-engineering campaigns, lower barriers for cyber criminals and adapt attacks in real time. Bowman said banks should maintain current asset inventories, phishing-resistant multifactor authentication, identity and access controls, and vulnerability identification and patch-management programs. She also called for employee training and periodic incident-response testing. The Federal Reserve tailors IT examinations to an institution’s risk profile and emerging threats and risks.
→ Action
Cybersecurity: Add AI-enabled social-engineering campaigns, automated vulnerability discovery and real-time attack changes to the next incident-response tabletop exercise.
Read article → from Federal Reserve
Nasdaq adds agentic capabilities to Calypso
For the treasury operations head at a bank
This makes trade-lifecycle automation a control decision, not just a productivity purchase. Your team should demand proof that sandboxing, live oversight and MCP connections prevent an agent from crossing data and approval boundaries.
Nasdaq launched an agentic AI operating environment within Nasdaq Calypso for capital-markets and treasury operations. Clients can use Calypso agents and connect proprietary AI infrastructure through a layer built on the Model Context Protocol, which links AI agents to software. The environment applies operational boundaries, live oversight and sandboxing with no external data retention, while agents can use Calypso trading, risk and collateral records plus enterprise, market and reference data. Nasdaq’s first capability is a natural-language assistant for platform data and documentation; it plans further governed agentic workers. The agents will run in Calypso’s Amazon Bedrock-based cloud environment, while on-premises clients can connect to them.
→ Action
Treasury operations: Test sandbox boundaries, live oversight, data-retention controls and MCP connections before linking proprietary AI to Calypso data.
Read article → from Finextra
General AI
Large language models & AI infrastructure
OpenAI introduces GPT-6.1 Sol
For a model-risk leader at a US bank
OpenAI’s claimed speed and affordability will make agentic deployments harder to defer. Treat the model’s Critical cybersecurity rating as a use-case gate: restrict privileged actions and require enhanced testing before production.
OpenAI introduced GPT-6.1 Sol, part of the latest model family in its GPT-6 series. OpenAI says Sol offers capabilities comparable to GPT-6 Astra, its most powerful model, with greater speed and affordability. Under OpenAI’s Preparedness Framework, OpenAI treats GPT-6.1 Sol as Critical in cybersecurity and High for biological and chemical capability. OpenAI applies the same safeguards stack used for GPT-6 Astra. Its evaluations took place in a research environment or through the API, which OpenAI says may produce slightly different output from production ChatGPT because of system prompts, tools and effort settings.
→ Action
Model risk: Require cybersecurity threat testing and approval before GPT-6.1 Sol production use.
Read article → from OpenAI
OpenAI rolls out Dots agents
For the COO of a regional bank
A tool that learns preferences, works 24/7 and reaches 4,000 apps turns routine automation into a standing delegation of authority. Your bank should not deploy it until identity, approval and audit controls cover every agent action.
OpenAI began rolling out Dots across its Pro, Business Premium and Enterprise plans. The agents run on GPT-6 Astra, have their own cloud computer, learn from feedback over time and can work toward user goals 24/7. Through plugins, Dots can connect to more than 4,000 apps and are available through ChatGPT, Slack and Teams. OpenAI says a Dot can learn a user’s preferences and standards; an early tester’s Dot prepared and sent an invoice after the user approved it.
→ Action
Identity and access management: Map least-privilege access, approvals and audit-log requirements before connecting any always-on agent to bank systems.
Read article → from OpenAI
Get this in your inbox every morning
Free · No spam · Unsubscribe anytime