A bank should treat an AI agent as a component of a bank-controlled process—not as an accountable employee, a substitute for governance, or a way to transfer responsibility to a vendor. Who is legally liable when an agent causes harm depends on the jurisdiction, the bank’s activities, the customer relationship, contracts and the facts. Current U.S. and European supervisory materials point to risk-based governance and controls, but they do not establish a universal agent-specific liability rule.
What current supervisory guidance says—and does not say
United States: model-risk guidance expressly excludes agentic AI
On April 17, 2026, the OCC, Federal Reserve and FDIC issued revised interagency model-risk guidance. It describes risk-based practices for model development and use, validation and monitoring, governance and controls, and third-party products. It is tailored rather than prescriptive, and the bulletin says noncompliance with the guidance alone will not result in supervisory criticism. Most relevant to banks over $30 billion in assets, it may also matter to smaller institutions in some circumstances.
As an Amazon Associate I earn from qualifying purchases.
The guidance defines the models within its scope around complex quantitative methods and expressly excludes generative and agentic AI. It is therefore not an agent rulebook. In a May 1, 2026 speech, Federal Reserve Vice Chair for Supervision Michelle W. Bowman likewise said generative and agentic AI fall outside the revised guidance, while other risk-management and governance practices are expected to support adoption. Her remarks describe a supervisory perspective, not a new binding requirement.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →United States: broader risk-management themes still matter
The U.S. Treasury’s financial-sector cybersecurity report discusses practical governance themes for AI-supported activities: assess risk and conduct due diligence before adoption; establish that a system suits its intended purpose; ensure adequate expertise and resources; test and validate; monitor performance; maintain an AI inventory; track issues and incidents; and apply relevant security, privacy, operational, fraud and resilience controls. These are themes in the report, not a bespoke legal checklist for agents.
Federal Reserve SR 20-24, an interagency paper on operational resilience, draws on operational-risk, continuity, third-party, cybersecurity and recovery disciplines. It is directed at specified large and complex firms, not automatically every bank. Its June 2, 2026 revision removed references to reputational risk in the attachment.
Europe: supervisory attention includes generative AI
The ECB’s 2026–28 supervisory priorities call for bank AI strategies that reflect both opportunities and risks, supported by robust governance and risk controls. The ECB says it will monitor AI generally and give targeted attention to banks’ generative-AI applications, while cooperating with relevant authorities on implementation of the EU AI Act.
ECB supervisory material also highlights explainability for decision-making, lifecycle monitoring and validation, change management, drift detection, escalation and remediation. For generative AI, it identifies concerns including provider concentration and vendor lock-in, data confidentiality and security, resilience, exit strategies, and legal and reputational exposure. These supervisory priorities signal areas of attention; they should not be presented as a single agent-specific rule.
Give an agent bounded authority inside an accountable process
Before deployment, define the process the agent supports and the consequences of an incorrect action. Map its permitted actions, accessible data, tools, external services and vendors. Decide which actions are reversible, which require human approval, and who can stop the workflow. The bank should also assess whether it has the expertise and resources to manage the system over its lifecycle, not just launch it.
During implementation, test both the agent and the surrounding workflow for the intended use. Validate the aspects that matter to the decisions or actions it supports; record versions and material changes; and maintain an inventory of AI uses along with issue and incident records. The Treasury report identifies these as governance themes. The revised interagency model-risk guidance also discusses validation, monitoring, governance, role clarity and third-party products for models in its scope, but it excludes generative and agentic AI.
In operation, monitor behavior and outcomes, set escalation thresholds, and define when the process pauses for review, suspension or rollback. Plan a workable alternative before a business function depends on the agent. ECB material emphasizes continuous monitoring, drift detection, escalation and remediation. Federal Reserve Bank of New York Chief Risk Officer Mihaela Nistor described a related failure pattern in a September 24, 2026 speech, delivered in her personal capacity: deployment can move ahead of governance’s understanding of decision concentration or resilience teams’ fallback planning. Her analysis is not a Federal Reserve requirement.
How an agent failure can reach customers or disrupt a bank
Unintended action or tool misuse
An agent may take an action outside its intended business purpose or authority. The cited materials do not establish agent-specific failure rates or a validated list of common failures. They do support controls such as defining intended use, testing, monitoring and tracking incidents. The practical question is whether the action was within an approved boundary and whether the bank could detect and contain an error.
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minuteDrift or changed behavior
Data, models, tools and workflows can change. Behavior that was acceptable at launch may become unsuitable as conditions shift. ECB supervisory material points to lifecycle monitoring, validation, change management and drift detection, with escalation and remediation when problems emerge.
Opaque decisions
If a bank cannot explain an agent’s behavior in terms meaningful to the decision it supports, it may struggle to assess, challenge or correct that behavior. An ECB supervisory speaker put the point this way: “If a bank cannot explain why an AI model behaves the way it does, in terms that are meaningful for decision-making, then it cannot truly control that model.” The statement appears in the February 24, 2026 speech “Technology is neutral, governance is not: AI adoption in the banking sector.”
Vendor dependency or disruption
An agent may rely on a foundation-model provider, cloud platform, external data, APIs and downstream services. Concentration, lock-in, confidentiality or security problems, provider outages and weak exit options can impair the bank’s ability to operate or recover. ECB material highlights these risks; the U.S. resilience paper also notes that third-party reliance can expose firms to operational risk.
Controls that lag deployment
Deployment can outpace governance, workforce readiness or fallback design. Nistor’s September 24, 2026 speech, given in a personal capacity, captures two possible gaps: “A process can become dependent on AI before resilience teams have designed a credible fallback,” and “A function can deploy autonomous agents before governance fully understands the resulting concentration of decision-making.” These are the speaker’s analysis, not supervisory requirements.
Who may be liable when a customer is harmed?
The cited sources do not settle liability for every bank, jurisdiction or incident. It would be too broad to say that a bank is always liable, that a vendor is liable, or that the agent itself bears responsibility. A legal analysis needs to account for the bank’s charter and activity, the customer relationship, applicable consumer, privacy, prudential and financial-services rules, contractual allocation, agency principles and the facts of the action that caused harm.
Best Value
The OECD’s 2024 comparative report gives one jurisdiction-specific example: in Israel, the licensed institution is liable to its client for harm following deployment of digital tools, including AI-related innovation, and cannot redirect the client to claim against a service provider. That example illustrates one country’s approach; it does not determine liability for a bank elsewhere.
Operational accountability is a separate question from the eventual legal allocation. A bank still needs owners for the process, controls for the agent’s authority, records of changes and incidents, monitoring, escalation and recovery planning. A contract with a provider may allocate obligations between the parties, but it does not by itself answer what a customer can claim under the law governing that relationship.
Quick Recap
Questions to settle before granting an agent authority
- Authority and reversibility: What actions can the agent take, and can each be reversed or stopped?
- Human review: Which decisions require approval, and what event triggers escalation?
- Explainability and evidence: Can staff understand the agent’s behavior well enough to make and document a decision?
- Monitoring and change: How will the bank detect drift, review material changes and respond to incidents?
- Third-party exposure: Which providers, data sources and services are dependencies, and what are the security, concentration and exit risks?
- Fallback and recovery: How will the bank keep the process running or restore it if the agent or a provider fails?
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Free tools Windows power users keep installed
One-click scans. No signup required.
