6 Aug 2026

The framing in most agentic marketing coverage is about efficiency: fewer manual workflows, faster content cycles, agents that adapt rather than fail. That's fine as far as it goes. But for anyone running a consumer credit business in the UK, the interesting question is not whether AI agents can repurpose content faster. It's what happens when an agent with write access to your marketing channels makes a decision your compliance team didn't anticipate.

Traditional marketing automation has a useful property that rarely gets credited: it's auditable by design. You mapped the workflow, you approved the rules, and when something goes wrong, you can trace exactly which condition triggered which action. Agents break that model. They determine their own steps toward an outcome, which means the path between your strategic intent and your customer-facing output is no longer fully documented in advance.

In consumer finance, that path is regulated territory. The FCA's Consumer Duty requires firms to demonstrate that communications are fair, clear, and not misleading. That's hard to evidence when the communication was shaped by a chain of agent decisions that weren't pre-approved, even if the final output looks fine.

This doesn't mean agentic approaches are incompatible with a regulated environment. It means the "humans approve the strategy and final output" framing, which appears in almost every agentic marketing guide, is necessary but not sufficient. You also need:

• Logging at the agent decision level, not just the output level
• Defined constraints on what tools an agent can invoke without human review
• A clear account of how the system behaves when it encounters an edge case

Credit brokers are already building loan origination systems with automated decision components, and the compliance infrastructure for those is reasonably mature. Marketing automation is treated as a different category, lower risk, less scrutiny. Agentic marketing probably closes that gap faster than most marketing or technology leaders have registered.

Rubrik's Agent Identity announcement is quietly one of the more important infrastructure moves in AI security this year. The core idea is simple: when an AI agent calls a tool, it gets a short-lived, scoped permission for that specific action, verified at runtime through Okta or Microsoft Entra ID. The agent doesn't hold standing privileges. It earns them, momentarily, for each operation.

For anyone building agentic workflows in consumer finance, this matters enormously. We are not far from a world where an AI agent inside a loan origination platform can query a credit bureau, update an application record, trigger a decisioning model, and send a customer communication, all without a human in the loop. That chain of actions, if running under a single broad service account, is a compliance and security nightmare waiting to happen.

The FCA's operational resilience framework already demands that firms understand and control what their systems can do. Extending that logic to AI agents is not optional. Regulators will eventually ask: who authorised that agent action, at what time, with what scope? "The model had access" is not an answer.

The identity gap in current AI deployments

Most teams deploying AI agents today are bolting them onto existing service accounts or API keys. The permissions are static, overly broad, and largely invisible to audit tooling. Rubrik is pointing at a structural problem the industry has been ignoring:

• Agents are actors, not just software, and actors need identity
• Short-lived credentials reduce blast radius when something goes wrong
• Runtime monitoring means you catch anomalous behaviour as it happens, not in a post-incident review

The MCP angle is worth watching too. Model Context Protocol is becoming the standard way agents connect to tools and data sources. Securing MCP connections at the identity layer is the right place to intervene.

The question for UK financial services leaders is how long they plan to run agentic systems on trust and hope before something goes wrong that lands in a Subject Access Request or a regulatory investigation.