The Fed is operationalizing a general-purpose AI stack across core back-office functions
Fed Governor Christopher Waller said the Federal Reserve is embedding a new general-purpose AI into daily operations to drive efficiency across payments, financial management, HR, and services provided to the U.S. Treasury. Separately, Governor Lisa Cook warned AI-driven productivity gains could create macro effects that challenge the Fed’s traditional policy toolkit. Net: the central bank is simultaneously deploying AI internally and signaling it expects second-order economic impacts.
Action: Benchmark your AI governance and operating-model maturity against what your primary regulator is building internally, then pre-brief your board on the productivity/labor-market implications the Fed is explicitly modeling. Prioritize controls for AI use in payments and finance operations (auditability, model risk management, vendor oversight) because these are now squarely in the Fed’s own transformation path.
Citi is building a dedicated AI-infrastructure banking franchise (and making direct AI bets)
Citi has formed an AI-focused Infrastructure Banking unit to capture advisory and lending tied to the data-center/compute buildout it estimates will require trillions in capital. The bank also disclosed a first investment in Japan’s Sakana AI, signaling it wants both balance-sheet exposure and ecosystem positioning, not just fee income.
Action: Stand up an explicit “AI infrastructure” coverage and risk playbook (power, land, permitting, supply-chain concentration, offtake/tenancy risk) before competitors lock in relationships. Tighten credit and concentration limits for data-center adjacencies (utilities, REITs, hyperscaler counterparties) because this theme is shifting from tech capex to core corporate/structured finance.
Compliance automation is consolidating: CUBE buys agentic AI risk-mapping firm 4CRisk.ai
Regtech CUBE acquired 4CRisk.ai, which positions agentic AI to map regulations to enterprise risks and controls. The deal signals buyers believe workflow-level “agentic” compliance (not just alerting) is now a defensible product wedge and that scale platforms want to own the mapping layer.
Action: Pressure-test your regulatory change and control-mapping processes against agentic tooling—especially where mapping is manual, slow, and audit-sensitive. Use consolidation as leverage in vendor negotiations (data portability, model explainability, liability terms) because platform lock-in risk is rising as suites expand via acquisition.
Anthropic is turning Claude into enterprise ‘agent plugins’—including finance workflows—raising the bar for in-tool automation
Anthropic launched new Claude “Cowork” plugins/connectors aimed at automating role-specific work inside existing enterprise software, with explicit packages for financial services use cases. This is a move from chat-centric AI to embedded agents that can execute multi-step tasks, challenging point SaaS tools that rely on human-in-the-loop workflows.
Action: Inventory the top 10 finance/risk/ops workflows where an agent could take action (not just draft text), then define the approval, logging, and segregation-of-duties controls required before rollout. Reassess SaaS vendor roadmaps and renewals in functions like research, underwriting support, and wealth ops because “agent inside the workflow” is now a first-class product direction from major labs.
OpenAI is publishing concrete scam TTPs it’s disrupting—useful indicators for bank fraud and cyber teams
OpenAI’s latest threat report details how criminals combine AI with websites and social platforms to run scams and other malicious operations, and it describes specific tactics the company disrupted. The key shift is operational: more of the “scam stack” is being industrialized with AI assistance, accelerating iteration and personalization.
Action: Feed OpenAI’s described tactics into fraud typologies, social-engineering training, and detection rules (synthetic personas, rapid content variation, multilingual lures, scripted outreach). Tighten controls on customer communications and call-center authentication because AI-assisted scam throughput will pressure existing step-up and dispute processes.
Accenture is adding Mistral to its ‘big three’ model lineup, signaling real enterprise demand for non-US LLM options
Mistral AI signed a partnership with Accenture, which already partners with OpenAI and Anthropic. This validates that large enterprise buyers are actively building multi-model strategies and that procurement, data residency, and sovereignty requirements are driving credible demand for European model providers.
Action: Accelerate a multi-model vendor strategy (commercial + open + regional) with clear workload placement rules (data sensitivity, latency, cost, residency). Use Accenture’s endorsement as a forcing function to revisit your model risk framework for third-party LLMs beyond the dominant US vendors—before business lines pick them up ad hoc.