Hope Sterling: Michael, hey — did you actually read the whole Mills Review or did you do what I did and immediately skip to the part that made you want to throw your laptop?
Michael C. Vincent: I read it twice. The second time more slowly, because the first time I kept stopping.
Hope Sterling: Wait, that's what I want to get into, because like, Sheldon Mills and the FCA put out this thing on July 6th, 2026 — and they're calling it the first regulator-initiated long-term AI review globally, which, fine, that's a big deal — but I keep getting stuck on the gap between what the review is ABOUT and what is already happening right now, like today, while we're recording this.
Michael C. Vincent: And what is already happening?
Hope Sterling: One in five UK adults — that's 11 million people — already willing to let AI make autonomous financial decisions. Not someday. Now. And more than a quarter of those people are trusting ChatGPT or Claude for financial advice with literally zero Consumer Duty protection underneath it. Zero.
Michael C. Vincent: Well. That number does not get easier the more times you say it.
Hope Sterling: It doesn't! So the question I cannot shake — and it's the thing this whole episode is really trying to figure out — is: if the Mills Review is meant to be foresight, why does it feel like the FCA is already four years behind the people it's supposed to be protecting?
Michael C. Vincent: That is exactly the right question. And I'm not sure the answer is flattering.
Hope Sterling: But like — that gap you're describing, that's the actual thing, right? Because I still don't think I fully understand what agentic AI even does differently. Like mechanically, what's the change?
Michael C. Vincent: Plain version: right now, a finance app suggests something and you tap yes. That tap is your consent, your moment of agency. Agentic AI removes the tap. The money moved before you picked up your phone.
Hope Sterling: Wait — it already moved?
Michael C. Vincent: You agreed to parameters, back at setup. The system operates inside them — continuously, not when you check in. That's what the Mills Review calls the shift from episodic to delegated. You're no longer a participant in your own financial life. You're more like — a beneficiary watching from the outside.
Hope Sterling: Okay that's — no, that framing actually breaks my brain a little. Because JP Morgan is already treating this as a CFO decision, not just a tech question — they said that out loud, publicly — which means the boardrooms have already made the leap. And we're sitting here with a 28-day consultation window?
Michael C. Vincent: The FCA Engagement Paper launched January 26th, 2026. Closed February 24th. Twenty-eight days. One hundred and forty written submissions total — from firms, consumer groups, academics, everyone.
Hope Sterling: And JP Morgan's legal team had those 28 days the same as, like, a lone academic trying to articulate a risk she's still discovering as she writes the paper.
Michael C. Vincent: That asymmetry is real. Now — Emma Banymandhub, CEO of The Payments Association, said explicitly that firms should treat agentic AI as an accountability and governance issue right now, not a 2030 problem. She's not describing the future. She's describing what's failing in the present, which makes the 28-day window feel — I'd put it carefully — like a photograph of who got heard loudest.
Hope Sterling: And the Senior Managers Regime — the whole accountability framework — it was built for a human making a visible decision. When the AI already moved the money, who exactly does the regime hold responsible?
Michael C. Vincent: That's exactly the crack in the foundation. The Senior Managers Regime assumes a human made a choice — a visible, reconstructable choice. You can point at it. You can ask 'why did you do this.' But when an agentic AI initiates a transaction inside agreed parameters, the senior manager didn't make that decision. They approved a system that made it. Which is — I mean, that's a genuinely different thing.
Hope Sterling: Wait, so the regime doesn't break — it just becomes decorative?
Michael C. Vincent: Ornamental was the word I kept landing on. It's still there. On paper, there's a name attached. But accountability needs somewhere to grip, and if the decision happened at machine speed inside a black box the senior manager may not be able to reconstruct — what are we actually holding them to?
Hope Sterling: And the FCA's answer to this is — no new rules. Just, like, Consumer Duty and the Senior Managers Regime, the existing frameworks, which were literally designed around a person who sat down and made a call. That's the answer? I'm not — okay, I'm not buying that.
Michael C. Vincent: The Mills Review does issue seven priority recommendations — monitoring autonomous models, establishing foundations for agentic finance specifically. That's not nothing. But there's no new rulebook. The review uses the words 'system-wide' and 'systemic' risk. Those are not casual words. Regulators reach for systemic when existing tools weren't built for the threat.
Hope Sterling: So they're naming the size of the problem with one hand and handing you a framework from 2016 with the other.
Michael C. Vincent: That is a sharp way to put it — yes.
Hope Sterling: And the Bank of England flagged this too — separately, the day after the Mills Review dropped, July 7th, Reuters reported it — they're out there saying AI poses growing risks to financial stability. So it's not like the FCA is alone in seeing the systemic dimension. Which honestly makes the 'existing tools are fine' position feel even more — I don't know, actually, no — it feels less like confidence and more like a holding pattern.
Hope Sterling: And just — flag for later — because the thing that makes all of this so much more complicated is that AI could genuinely close the advice gap for millions of people who've never had access to a financial advisor. But those same systems create a risk concentration problem that has no historical precedent. We'll get there.
Michael C. Vincent: That closing is exactly where I want to push — because the advice gap isn't theoretical. Picture a delivery worker in Birmingham, 28 years old, never earned enough to justify a human advisor's hourly rate. The Mills Review identifies this as one of four structural shifts AI will drive by 2030 — the evolution of consumer journeys. Agentic AI could give her genuinely personalised guidance for the first time in her life. That is real. That is the promise.
Hope Sterling: No, and that part actually — I mean, I feel that, that's the part that makes me want to defend this whole thing.
Michael C. Vincent: But the same model serving her is serving three million others. One bias baked into training data, one security breach — harm doesn't travel person to person anymore. It propagates simultaneously. Retail finance has never faced that shape of failure.
Hope Sterling: Wait — simultaneously? Like, not a wave, just — everyone at once?
Michael C. Vincent: Everyone on the same model, yes. And the Bank of England flagged exactly that architecture of harm — reported July 7th, 2026, the day after the Mills Review published. They didn't coordinate that timing, which actually — I mean, that's what makes it land harder. Two independent institutions, same week, same conclusion about systemic fragility.
Hope Sterling: One day apart. That's not a coincidence you can just walk past.
Michael C. Vincent: The Mills Review does name this — transformation of firm operations, reshaping competition, amplified financial crime and cyber risk, the consumer journey shift. All four of those structural changes converge on the same chokepoint: concentrated model dependency. And one of the seven recommendations actually calls for AI-enabled supervisory models — which, you see, that's the FCA quietly admitting their current supervisory toolkit wasn't built for this either.
Hope Sterling: Hold on — they're recommending AI to supervise AI? Because their human-scale tools can't keep up with — okay that's actually the most honest thing in the whole document.
Michael C. Vincent: That's the genuine insight buried in it. The democratisation is real. The risk concentration is real. And the FCA is — however imperfectly — trying to hold both.
Hope Sterling: But that's actually — wait, the Mills Review is literally the first work of its kind by a regulator globally, like, they said that, that's their own words — but being first and being enough are just... not the same thing? And the FCA explicitly declines to write new AI-specific rules, even after naming system-wide risks. So does Sheldon Mills actually believe Consumer Duty can hold this? Or is the 2030 framing just — a way to be remembered as the regulator who saw it coming, whether or not they stopped it?
Michael C. Vincent: I don't know. And I mean that plainly — I genuinely don't know the answer. That question might not have one yet.
Hope Sterling: Yeah. That's kind of where I am too. Which is uncomfortable but also — I think that's the honest place to stop.
Michael C. Vincent: It is. Thank you for pushing on it as hard as you did.
Hope Sterling: Genuinely, same. This one stuck.
