BankingNewsAI Daily Brief ·
State bank regulators issue an AI oversight road map for examiners
Banking AI
Financial institutions & fintech technology
State bank regulators give examiners AI oversight road map
For a chief compliance officer at a state-chartered bank
State examiners now have a ready-made list of AI evidence to request, even where no new rule exists. Your bank must decide whether its AI governance can withstand that review or expose unmanaged vendor and consumer-risk gaps.
The Conference of State Bank Supervisors released an AI supervisory framework for examiners of state-chartered banks and state-licensed nonbank financial institutions. The framework gives examiners questions, document requests and criteria for deciding when an AI system needs closer scrutiny, though each state agency decides whether and how to use it. Examiners may request AI policies, use-case inventories, risk assessments, management reports, vendor contracts and testing records. For customer-facing systems, they may seek chatbot transcripts, notices and other AI-assisted communications. The guide asks about ownership, risk ranking, control updates, and, for agentic AI, action limits, human checkpoints, logs, reversibility and stop controls.
→ Action
Compliance: Build an examination-ready AI inventory linking each use case to owner, risk assessment, vendor contract, testing evidence and customer communications.
Read article → from PYMNTS
Feedzai launches Farol for fraud analysis
For the fraud operations head at a regional bank
Farol turns analyst approval limits and audit evidence into operating controls, not governance paperwork. Before allowing rule suggestions or SAR drafts into production, define who can approve them, what evidence they need and what must be retained.
Feedzai launched Farol, an AI agent embedded in its RiskOps Studio for bank fraud operations. The agent provides conversational intelligence, rule-performance analysis, case summaries and product guidance within the workflow analysts already use. Its Risk Strategy skill identifies rules that generate noise without catching fraud and recommends threshold changes, while its Investigations skill retrieves and summarizes alert data; Feedzai says this reduces alert handling time by 20%. Farol can also draft Suspicious Activity Reports up to 12 times faster, according to Feedzai, and operates within each financial institution’s environment so its information and insights stay inside the customer’s estate.
→ Action
Fraud operations: Test Farol case summaries, rule recommendations and SAR drafts against analyst approvals; retain an audit trail for each recommendation and production change.
Read article → from Feedzai
LoanPro launches payments, disbursement and AI servicing tools
For the consumer-lending COO at a regional bank
AI servicing procurement now turns on whether agent actions inherit employee controls. Do not accept a vendor’s audit-trail claim without testing role access, action attribution and servicing permissions end to end.
LoanPro launched DirectPay, Beyond Credit and LoanPro MCP for payments, loan disbursement and AI-assisted servicing. DirectPay processes about $4 billion a month in loan repayments and disbursements, and lets loan officers view a payment’s lifecycle, failed-payment reasons and disputes on one platform. Beyond Credit is a fee-free virtual Mastercard that lets borrowers access installment-loan funds through Apple Pay or Google Pay and lets lenders restrict spend by merchant, location or merchant category code. MCP connects lender-selected AI agents and models to LoanPro servicing workflows under the same permissions, compliance guardrails and audit trails used for employees. Each agent action is recorded under the employee using the agent.
→ Action
Model risk: Test whether AI-agent actions enforce employee role permissions and record the responsible employee across servicing and payment workflows.
Read article → from Finovate
General AI
Large language models & AI infrastructure
Microsoft Foundry expands model choice and voice agents
For the CTO at an Azure-based bank
Foundry lowers the cost of changing models, so your advantage will come from the test suite and approval rules around each swap, not the model you chose first. Treat quality, latency, and cost testing as a production control.
Microsoft expanded Foundry with GPT-6 models from OpenAI, Claude Opus 5.5 from Anthropic, and voice agents in Foundry Agent Service. Foundry lets teams evaluate models against their own data and workloads, compare quality, latency, and cost, and put selected models into production without rebuilding around one provider. Voice agents are in public preview and support prompt and hosted agents, with GPT Realtime, Azure Realtime, MAI, and bring-your-own models. They support more than 80 languages and 140 locales, reuse existing tools, enterprise knowledge, and governance, and deploy to web, Teams, Teams Phone, and Twilio telephony. Foundry also uses production traces to evaluate and refine instructions, skills, tools, and model choice.
→ Action
AI platform engineering: Create workload tests that compare approved Foundry models on quality, latency, and cost before production changes.
Read article → from Microsoft
BNP Paribas forges AI partnership with Google Cloud
For the CIO of a corporate bank
Do not let one model provider set your agent roadmap. Credit-memo agents demand model choice, control and evaluation before wider use in sales or trading.
BNP Paribas signed a five-year deal to expand access to Google Cloud infrastructure and AI tools, including Gemini Enterprise and Gemini models. Its Corporate & Institutional Banking unit will first deploy purpose-built agents for targeted workflows, including corporate credit-memo preparation. The bank plans further agent uses in sales, trading, research and structuring. BNP Paribas will integrate Gemini models into LLM@CIB, its generative AI assistant used by more than 65,000 employees. The bank also renewed its Mistral AI partnership for three years in May, focused on cyber defenses, as part of a multi-cloud and multi-model approach.
→ Action
AI governance: Set model-selection, evaluation and control requirements for credit-memo agents before approving production use.
Read article → from Finextra
Get this in your inbox every morning
Free · No spam · Unsubscribe anytime