Identity as the Regulatory Shield for Agentic Banking

For years, banks have treated artificial intelligence as a passive analytical tool. From rules-based reconciliation bots in the back office to machine learning models for credit risk, we have been focused on prediction. We are now entering the fourth wave of AI: Agentic AI. Unlike the supervised copilots that preceded them, these agents act as autonomous entities capable of executing transactions, extending credit, and binding policies.

For banking, this shift from analysing to doing is revolutionary. However, it also creates a significant governance gap. If a bank’s agent initiates a £50,000 transfer, who is accountable? The regulatory answer is clear: autonomy must not dilute accountability. To thrive in this new landscape, banks must treat the identity fabric as the essential control plane for autonomous banking.

The New Control Loop: Autonomous Fraud Defense

Fraud remains a critical operational risk for banks, particularly with the surge in Authorised Push Payment (APP) fraud, which is a primary target of current FCA and PSR regulatory attention. Static, rules-based fraud triage is simply too reactive.

The target state for banking is a continuous, identity-anchored control loop. In this architecture, an authenticated “business-side” fraud agent monitors payments in real-time. When it detects an anomaly, it doesn’t just block the transaction; it initiates a bilateral check with a customer-owned agent. Through delegated authorisation, the bank confirms intent before the funds move. This is the difference between a system that detects a crime after the fact and one that prevents it through verified, identity-based orchestration.

Transparency in Credit Underwriting

Credit underwriting has long been a fragmented, opaque process often prone to bias and delays. Agentic AI offers a path to digital-speed underwriting, but regulators - particularly under the EU AI Act - rightly demand transparency and explainability.

By utilising an identity fabric, banks can ensure underwriting agents operate within a chain of trust. A bank’s underwriting agent can connect with credit bureaus via the Model Context Protocol (MCP), a common language that allows agents to discover and validate one another securely. Through delegated rights, the agent gathers only the data authorised for that specific application, and policy-based access control (PBAC) ensures the decision adheres to internal risk limits and jurisdictional lending rules. Every data exchange and reasoning step is immutably recorded, providing the evidence base that auditors demand.

Compliance as an Architectural Guarantee

European and UK regulations - such as DORA and the FCA operational-resilience rules - are setting the stage for acceptable AI use. Banks can no longer treat compliance as a post-hoc reporting exercise. Identity serves as the operational embodiment of this compliance.

By embedding identity-led principles such as least privilege permissions, just-in-time access, and human-in-loop controls - banks turn regulatory requirements into automated, enforceable code. When an agent initiates a high-impact action, such as a large fund transfer, the system automatically triggers a four-eyes review. Oversight occurs at the speed of automation, not after the risk has materialised.

Authenticated, Authorised, Observed, and Attributable

Autonomous banking is not merely a technological upgrade; it is a fundamental change in how banks operate, engage clients, and manage institutional risk. As we scale, we must move away from point-to-point integrations and embrace a unified identity fabric that treats every AI agent as an accountable, identifiable entity.

For banking leaders, the goal is simple: ensure that every autonomous decision is authenticated, authorised, verified, and auditable. By doing so, we transform compliance from a constraint into an enabler, building the trust required to innovate safely in the agentic era.