Banking AI-DLC Identity | Moring
AI-DLC identity in Banking means every coding agent gets its own governed identity, scoped like any user with elevated access. Agent credentials carry defined permissions, approvals and audit records, so banks can attribute each AI-generated change to a specific actor. This identity model supports SR 26-2, SR 11-7, FFIEC, EU AI Act and DORA oversight of AI-driven development.
Short answer
AI-DLC identity in Banking means every coding agent gets its own governed identity, scoped like any user with elevated access. Agent credentials carry defined permissions, approvals and audit records, so banks can attribute each AI-generated change to a specific actor. This identity model supports SR 26-2, SR 11-7, FFIEC, EU AI Act and DORA oversight of AI-driven development.
How it works
In AI-DLC, agents do the executing, so the identity question moves to the front. A coding agent that writes code, opens changes and touches pipelines behaves like a user with elevated access, and banks already know how to govern that kind of principal. The AI-DLC governance harness starts by giving each agent a distinct identity with permissions scoped to granted models and tools, so attribution of every action is possible from the first commit.
The industry framing supports this. Commentary in Asian Banking and Finance argues that as AI agents move from assisting employees to acting autonomously in payments, lending, fraud management and customer servicing, they need to be governed the way banks govern any user with elevated access, not as software components that behave predictably once configured. Lloyds Banking Group's security leadership made the same point in practice: what an agent can do, and what evidence proves it did only that, are design decisions made before deployment.
Scoping is the core control. The agent identity is granted access only to the repositories, environments and tools its approved use case requires, and that scope lives outside agent code where the agent cannot rewrite it. AWS's financial services guidance on AI model governance and access controls makes the same demand: comprehensive access controls, model lifecycle management and continuous oversight, with approval workflows and change management ensuring review before models are deployed or modified.
Identity also carries the evidence chain. When the agent identity is the access control, the record of which identity ran which task, with whose approval, under which ticket, doubles as the audit trail examiners ask about under SR 26-2 and SR 11-7. FFIEC examination expectations for access governance and DORA-style resilience reporting both become easier to satisfy when the attribution record is produced automatically by the development process.
The harness approach makes identity portable across agent platforms. AWS describes the governance harness as agent-agnostic, working with Kiro, Claude Code or OpenAI Codex, which means a bank can define one identity model and apply it to whatever coding agents its teams adopt. Deloitte's work on agentic governance for banks reaches a compatible conclusion: lifecycle-managed specifications and automated control configuration are needed when institutions deploy agents at scale.
For teams starting now, the sequence is practical: register each agent as a named principal, attach its approved use case and owner, scope its credentials, route its consequential actions through approval gates, and log everything to the audit trail. Identity done this way converts AI-driven development from an ungoverned actor into an accountable one, which is the condition regulators in banking actually examine.
Frequently asked questions
Why do AI coding agents in banking need their own identities?
Because an agent that writes and deploys code acts like a user with elevated access, and banking controls are built around attributing actions to named principals. Giving each agent a distinct, scoped identity means every commit, pipeline run and deployment can be traced to a specific actor with a defined owner. Without that identity, banks cannot answer examiner questions under SR 26-2 or SR 11-7 about who or what made a change.
How should agent permissions be scoped in AI-DLC?
Scope each agent identity to the repositories, environments, models and tools its approved use case requires, and keep that configuration outside agent code so the agent cannot rewrite it. AWS financial services guidance calls for comprehensive access controls, model lifecycle management and continuous oversight. In practice, permissions follow the use-case specification, and any expansion of scope requires a fresh approval, keeping least-privilege discipline intact across the lifecycle.
What did Lloyds Banking Group say about agentic AI controls?
At Infosecurity Europe, Lloyds Banking Group's security team described putting agentic AI into production as an engineering problem to be designed, constrained and tested at scale. Running 23 million customer accounts and billions of log lines a year, the bank treats decisions about what an agent can do, and what evidence proves it did only that, as concrete design choices made before deployment rather than abstract risk debates.
How does agent identity support audit and resilience requirements?
When the agent identity is the access control, the system automatically records which identity performed each action, under which ticket and approval. That record is the audit trail FFIEC examinations and SR 26-2, SR 11-7 oversight look for, and it supports DORA-style resilience reporting by showing exactly which automated principals touched critical systems. Attribution produced by the process itself is stronger evidence than logs reconstructed after an incident.