The Agent Did the EDD. Who Signs It?

A chatbot that answers a compliance question is one thing. A system that goes and does the compliance work is another.

The second kind pulls the adverse media, queries the corporate registry, drafts the EDD write-up and flags the file for review. Several steps, no person in the loop until the end. It is already in production at a growing number of institutions, and not as a pilot either — it runs the first pass on real case volume. The capability is moving fast.

The question of who answers for it when it gets something wrong hasn’t moved at all. Worth being precise about why it can’t.

Agentic doesn’t mean autonomous in the sense that matters

“Agentic” describes execution. Multiple steps, tool use, limited human intervention mid-task. It says nothing whatsoever about who is answerable for the output, because in UK financial services that question was never a function of who, or what, did the work.

Under the Senior Managers and Certification Regime every prescribed responsibility sits with a named senior manager. Full stop. Drop an agentic system into that function and no gap opens in the org chart for the AI to fill. The senior manager who owns customer due diligence still owns it, whether the first-draft research came from a graduate analyst or from a language model chaining together six tool calls.

The system executes the task. It cannot hold the responsibility. The whole architecture was built around a named accountable person rather than a process, and a process is all the agent is.

This is worth stating plainly because a lot of vendor material implies otherwise — that sufficiently capable tooling somehow absorbs the risk. It doesn’t. It relocates where the evidence trail needs to point, and it changes what the senior manager has to be able to demonstrate they were overseeing.

Where the FCA’s approach actually lands on this

The FCA has been explicit, including from its chief executive as recently as December 2025, that AI-specific rules are not coming. The reasoning given is that the technology turns over on a three-to-six-month cycle, too fast for bespoke rules to stay current, and that existing frameworks already cover the ground that matters: SMCR, Consumer Duty, model risk management expectations. The AI Lab and the AI Live Testing pilot exist to let firms test deployments collaboratively with the regulator, not to hand down a new rulebook.

So there is no AI compliance checklist waiting to be published that will tell you how to govern an agentic EDD tool.

The obligations that already applied are the ones that apply. Can you show the output was overseen? Can you show a person made the judgment call? Can you show somebody reviews the process periodically to check it is actually working? Those questions land on the agentic version of the task exactly as they landed on the manual one. Firms holding off on formalising oversight until AI-specific guidance arrives are waiting for something the regulator has now said twice isn’t coming.

The failure mode that’s genuinely new

None of which means nothing has changed.

Agentic systems carry a risk a human researcher doesn’t, at least not in the same shape: fluent, confident and wrong. A junior analyst who isn’t sure of something hedges, or flags it, or leaves it out — uncertainty shows up in the writing. Ask a language-model-driven agent to summarise a source and it produces the summary in exactly the same confident register whether the source says what the summary claims or not.

The failure doesn’t look like a gap. It looks like finished work.

That changes what oversight has to check for. Reviewing agentic output for completeness is no longer enough, because completeness is the one thing these systems reliably deliver. Someone has to spot-check the underlying claims against the actual sources, sampling in proportion to risk, precisely because the failure is unobtrusive by the nature of how the text gets generated. A due diligence file that reads well and looks thorough has only ever been evidence that it is fluent. Agentic tools are exceptionally good at fluent.

What this means for a compliance function building toward this now

Three things, and all of them before an agentic tool touches live casework rather than after.

Name the accountable senior manager for that specific function, explicitly and in writing, at the point of deployment. Not once something has gone wrong.

Build the sampling-based verification step into the process design from day one, sized to the risk of the task. Output quality is not a proxy for output accuracy and never was.

And keep a clear, demonstrable line between what the agent drafted and what a human reviewed and approved. Not because the technology is unreliable. Because in a regime built entirely around named human accountability, that line is the only thing that was ever actually being tested.