Three critical gaps in the FCA’s vision for the future of financial technology
The FCA’s Emerging Technology Horizon Scan 2026 is a genuinely useful document. It maps three convergent futures: personalised AI sitting between us and our money, synthetic fraud at industrial scale, and programmable finance rewiring the plumbing underneath all of it. The work is careful and the scenarios are plausible. The authors are explicit that this is foresight, not prediction or guidance.
But foresight that stops short of the policy question leaves the most important part unbuilt. Here are three gaps I keep coming back to.
01. Who builds the guardrails, and with what authority?
The scan is candid about the risks, for instance, proxy agents quietly favouring the firms that fund them. It calls out the potential for “Dark patterns” aimed not at people but at the recommendation logic of the AI acting for them. Without control of the logic, the system could become a multi-agent environment that drifts into spoofing or collusion with no human in the loop. This is a world that moves at machine speed, with thousands of transactions per second harming real-world assets before anyone notices.
What it doesn’t do is ask how regulation itself becomes part of the answer. There’s no discussion of whether existing rules even reach machine-to-machine negotiation, no view on whether a proxy acting for a consumer owes that consumer a duty of care, no thinking on how you supervise a market whose primary participants are software agents talking to other software agents. The Consumer Duty was written for firms dealing with humans. What happens to it when the “consumer” the firm interacts with is an algorithm?
The analysis is careful; the gap is that it raises a sharp question and then steps back from answering it. A horizon scan can reasonably say “this is the regulator’s job to work out next.” But it should at least say that out loud, because the guardrails won’t appear on their own, and the firms building these agents are moving faster than anyone drafting rules for them.
02. The proxy economy assumes people want to hand over the keys.
How much of that is true? History perhaps says otherwise.
The scan describes a staged slide into delegation: assistants, then advisory, then “do-it-for-me,” where your agent negotiates prices, moves your money and reallocates investments while you skim a periodic summary. It compares the consent involved to clicking through cookie banners.
I think this badly underrates how much people resist handing over control of their money. We click through cookie banners because the stakes feel like nothing. People do not feel that way about their savings, their mortgage or their pension. Trust in financial institutions is already low, and the thing being asked here is far larger than tolerating a tracking cookie. It’s authorising a system to spend on your behalf while you look away.
Plenty of evidence points the other way from a frictionless transition. People abandon perfectly good robo-advisers the moment markets wobble. They keep money in current accounts earning nothing, rather than move it, partly out of inertia but partly out of a real reluctance to let go. A lot of us check our banking app not because we need to act but because watching is how we feel in control. “Do-it-for-me” asks people to give up exactly that, and the report treats the handover as a smooth gradient when it’s more likely to be a wall many people simply refuse to climb.
That matters for the scenarios. If adoption is lumpy and contested rather than gradual and universal, you don’t get one proxy economy. You get a divided market: a minority who delegate fully, a large middle who delegate the boring bits and guard the rest, and a sizeable group who opt out entirely and may end up paying more for the privilege of staying human. That resistance is what shapes the outcome, and the report treats it as a smooth gradient when it may be the most decisive variable of all.
03. Quantum computing is entirely missing.
The programmable finance chapter is built on a specific promise: that cryptography makes the new plumbing trustworthy. Tokenised assets, smart contracts, atomic settlement across sovereign ledgers, “the large-scale automation of trust.” Much of it rests on cryptographic guarantees.
Quantum computing is one foreseeable development that threatens those guarantees directly. A sufficiently capable quantum computer breaks the public-key schemes, such as RSA and elliptic-curve cryptography, that authorise transactions across these systems. The migration to post-quantum cryptography is already underway, with standards finalised by NIST in 2024 and the Bank of England publishing a paper in October 2025.
However, treating quantum only as a threat also misses the other half of it. The same machines could deliver real gains in finance, from faster simulation for pricing and risk to sharper portfolio optimisation and better fraud detection. Some of that is years off and some of it is oversold, but the trajectory points to capabilities that will eventually matter commercially. The harder question is who gets them. Quantum capability tends to concentrate, because the hardware is scarce, the skills to use it are in short supply, and the real advantage comes from having proprietary data to apply it to and from getting there first. Cloud “quantum-as-a-service” lowers the barrier to entry, but renting machine time is not the same as having the people to exploit it, or the head start that early access brings. If the firms best placed to harness quantum first are also the largest and best funded, the likely result is a widening gap in pricing, risk and security capability, which is the sort of concentration the report worries about elsewhere.
That silence is what stands out. The Bank of England’s own October 2025 paper put quantum alongside AI and DLT as one of three technologies set to reshape UK financial services. The FCA scan covers the same sector over the same horizon, and its third pillar is explicitly about the cryptographic rewiring of financial infrastructure, yet it does not mention quantum once. Nor does it apply to quantum the worry about concentration that runs through the rest of the report.
“We are excited to collaborate with KRM22 to bring our advanced sentiment analytics to a broader audience within the capital markets ecosystem. Together, we are enabling firms to unlock new value from unstructured data and strengthen their risk management capabilities”
-Andrew Simpson, CEO & Founder at SIGMA AI
Where this leave us
None of this is a reason to dismiss the scan. It’s a reason to read it as a starting point rather than a finished map. All three gaps point to the same underlying weakness: the document is clearer on what these technologies can do than on what will decide whether, how, and for whom they actually do it. Who writes the rules when the participants are algorithms? Will people actually accept the bargain being offered? And underneath all of it, conspicuously absent, is quantum computing, which touches both the cryptography the chapter depends on and who ends up holding the advantage, yet the scan never names it.
Those three gaps seem to me more decisive than any of the capabilities themselves. Curious whether others reading it landed in the same place.
