Fed Governor Barr Warns Agentic AI Poses Control and Accuracy Risks for Banks in Singapore Speech

Financial regulators and commercial banks are drawing hard defensive boundaries around automated capital movement to avert destabilizing balance shifts and unverified transaction liabilities.

• Regulators flagged autonomous execution risks: Federal Reserve Governor Michael S. Barr warned at the Singapore Fintech Festival that agentic AI presents distinct systemic challenges in credit assessment, fraud surveillance, and trading when software agents perform multi-step executions without continuous human oversight.
• Major institutions blocked consumer bots: By October 2026, JPMorgan Chase, Citigroup, Charles Schwab, and Wells Fargo implemented operational barriers against third-party AI agents attempting direct account actions.
• Terms shifted financial accountability: Wells Fargo enacted updated customer liability terms effective November 2026, explicitly transferring financial loss liabilities to users if unauthorized actions stem from third-party AI agents granted access to their accounts.

• Guidance gaps created an oversight void: Interagency regulatory guidance issued in April 2026 excluded generative and agentic architectures from traditional model risk management frameworks, leaving individual banks to construct independent defense perimeters.
• Deposit velocity introduces balance-sheet stress: Industry projections from Apollo Global Management estimate a $47 billion annual revenue cost to U.S. banks if consumer-directed agents continuously sweep low-interest operating balances into higher-yielding external vehicles.
• Bank valuations reacted to tech-led automation: Commercial bank equities declined following Meta's rollout of agentic execution tools, accelerating defensive interventions by institutions balancing internal AI deployment against third-party access risks.

The tension between institutional banking balance sheets and external software agents is fundamentally an incentive mismatch over transaction execution, speed, and financial liability.

When generative software advanced from conversational prompts to agentic architectures capable of completing complex, multi-step actions across digital environments, consumer and enterprise users gained the ability to delegate financial agency. Instead of requiring human intervention to execute payments, rebalance investment accounts, or shop for competitive yield spreads, agentic AI operates programmatically. It continuously interprets user goals, accesses digital interfaces, and executes financial transfers autonomously.

For deposit-taking institutions, this shift challenges standard liquidity and risk-underwriting models. Traditional retail and commercial banking models rely on deposit inertia: customers routinely leave substantial balances in zero-yield or low-interest checking and operating accounts. When third-party autonomous agents continuously query real-time market returns and route funds instantly toward optimized alternatives, deposit velocity spikes. Apollo Global Management calculates that automated sweeps run by external agents represent an estimated $47 billion annual cost to U.S. lenders through lost net interest margins and higher funding costs.

Simultaneously, agentic systems weaken standard verification architectures. When an autonomous tool initiates a transaction via consumer-provided credentials, banks struggle to distinguish between genuine user intent, algorithmic hallucinations, and malicious exploits. Governor Barr's November 2025 Singapore address identified this specific operational friction, noting that financial workflows demand controlled, explainable, and fully replicable execution. Agentic systems operating across external codebases do not inherently satisfy banking audit standards.

Because April 2026 interagency supervisory rules excluded generative and autonomous models from conventional model risk frameworks, financial institutions bore the full operational risk of unverified transactions. Lenders responded not by pausing internal automation, but by walling off customer accounts from outside bots.

By late 2026, JPMorgan Chase, Citigroup, Charles Schwab, and Wells Fargo moved to restrict or block external agents from accessing their interfaces. Wells Fargo established explicit liability boundaries in November 2026, stating that customers bear the direct financial loss for unauthorized executions initiated through external agents. Banks are adopting internal agentic workflows to increase their own productivity while shutting out consumer-facing autonomous software to protect balance sheets and legally isolate transaction liability.

• Corporate treasurers and private wealth managers: Face abrupt interruptions when deploying independent automation to optimize working capital across accounts at institutions restricting external API and screen interactions.
• Retail bank depositors: Lose automated yield-maximization tools as lenders enforce terms of service that assign full financial liability to the customer for bot-facilitated transactions.
• Third-party fintech and agent developers: Confront immediate operational roadblocks as proprietary protocols, anti-bot shields, and credential restrictions block access to major institutional customer portals.
• Compliance and operational risk officers: Must manage heightened audit inquiries from the Fed, OCC, and FDIC focused on shutdown capabilities, auditability, and monitoring of authorized internal agents.

• Interagency supervisory requirements: Watch for updated formal guidance from the Federal Reserve, OCC, and FDIC establishing mandatory shutdown switches and audit trails for bank-approved AI systems.
• Standardized liability litigation: Monitor customer-led legal challenges contesting whether banks can unilaterally assign liability to account owners when third-party agents execute disputed payments.
• API access and interface disputes: Track whether independent fintech aggregators bring regulatory complaints against Tier-1 banks for anti-competitive behavior under the guise of security protections.