Demis Hassabis published a manifesto this morning. The co-founder and CEO of Google DeepMind, a Nobel laureate in chemistry, and by most measures the person closest to the frontier of artificial intelligence, set out his view on where the technology is heading and what ought to be done about it. The essay covers AGI timelines, existential risk, and the usual menu of civilisational stakes. But the detail that stopped me was more specific.
Hassabis wants a FINRA for AI.
That is, he wants to model the governance of frontier artificial intelligence on the Financial Industry Regulatory Authority, the self-regulatory body that polices Wall Street under SEC oversight. Industry-funded. Independently governed. With the power to test, approve, and if necessary slow down the release of the most capable AI systems in the world. He could have reached for any regulatory precedent. He chose ours.
I work inside the UK mortgage market, where regulation is not a theoretical exercise. The FCA, the PRA, Consumer Duty, MCOB, the Senior Managers and Certification Regime — these are the frameworks that determine how we give advice, how we evidence it, and what happens when we get it wrong. They are familiar enough to be unremarkable. So there is something quietly interesting about one of the world’s foremost scientists arriving at the same structural conclusion that financial services reached decades ago: when the stakes are high and the technology is complex, self-regulation by informed practitioners, with independent oversight, is the least bad option.
The article is notable for its candour. “Nobody in the world knows for sure what is going to happen from here,” Hassabis writes, “and even the experts disagree.” He believes AGI, a system that exhibits all the cognitive capabilities of the human brain, is “probably only a few short years away.” He describes us as standing in “the foothills of the singularity” and estimates the impact at ten times the Industrial Revolution, at ten times the speed.
There is a sentence in the piece worth pausing on. “We’ve essentially found a way to make sand think.” From most people, that would be showmanship. From the person who built AlphaFold, predicted the structure of virtually every known protein, and won a Nobel Prize for it, it is closer to a factual observation.
The question for anyone working in UK financial services is what, if anything, this has to do with them. The answer, I think, is more than is immediately obvious.
The FCA said just last week that AI will “relocate intermediation, not remove it.” The phrasing is deliberate. The regulator is not predicting the end of mortgage brokers. It is observing that the nature of what brokers do, and where the value sits in the advice chain, is likely to shift. AI agents, platforms, and data layers will absorb more of the process. The human adviser’s role tilts away from product selection and towards judgment, empathy, and contextual understanding.
The parallel with Hassabis’s framework is worth noting. His proposed Standards Body would test frontier AI models before release, much as our regulatory framework requires advice processes to be evidenced before, during, and after the point of sale. The underlying logic is the same: when the consequences of failure are severe, you build structured oversight rather than relying on the market to self-correct.
The FINRA model in particular is interesting. FINRA is not a government agency. It is a self-regulatory organisation, funded by the industry it oversees, with a board that includes independent experts alongside practitioners. Hassabis is proposing something similar for AI: a body funded by the frontier labs, governed by an independent board of technical experts, with the authority to require pre-release testing and, if the situation warrants it, to coordinate a slowdown. For anyone who has spent time inside a mortgage network’s compliance structure, the architecture is recognisable.
The mortgage market already lives with the tension Hassabis is trying to solve: the commercial pressure to move fast set against the regulatory requirement to slow down, to evidence, to protect consumers from harm. These are not forces that can be permanently reconciled. They are managed, continuously, by people who understand both the technology and the rules. Hassabis appears to have concluded that AI needs the same arrangement.
The practical question for mortgage and protection professionals is not whether AI will change the industry. It already is. The question is whether we are preparing for a world where AI handles the sourcing, the admin, and the compliance documentation, and frees us to do the work that genuinely requires a human being. If Hassabis is even roughly right about the timeline, that world is not a decade away.
I use AI tools daily in my consulting work. Over the past eighteen months, I have watched them improve from clumsy novelties to genuinely useful working tools. They do not replace thinking, but they compress the time between a question and a well-researched answer to a degree that would have seemed implausible two years ago. When Hassabis describes the current moment as “pivotal,” it matches what I observe at the level of day-to-day professional work, not just in the abstract.
The FCA’s position is evolving in the same direction. It has said it will not introduce AI-specific regulation, preferring to apply existing principles-based frameworks to new technology. But it has also signalled that guidance on audit trails and human-in-the-loop protocols is coming, and it is actively encouraging brokers to adopt AI through its mortgage market reforms. The tone has shifted. The regulator is not merely tolerating AI in advice. It is encouraging it.
Hassabis closes his manifesto with a line that is easy to skim past but harder to dismiss: “The future is not yet written.” For an industry that sometimes feels as though its future is being drafted elsewhere, by lenders, aggregators, technology vendors, and regulators, that is worth considering. The mortgage professionals who engage with AI now, who learn its capabilities and its limitations, and who build it into their processes while retaining the human judgment that clients value, will have more say in what comes next than those who do not.
A Nobel laureate has told the world that nobody knows what happens from here. That is not a comfortable statement. But it might be a useful one.
Research notes
Demis Hassabis, “A Framework for Frontier AI and the Dawning of a New Age,” published 14 July 2026 on Substack, LinkedIn, and X. Full text at demishassabis.substack.com.
Axios exclusive interview with Hassabis by Mike Allen, Zachary Basu, and Madison Mills, “Google DeepMind’s Demis Hassabis calls for U.S.-led global AI watchdog,” 14 July 2026. Additional context on White House talks, the proposed timeline (“before year-end”), and the Anthropic Mythos incident.
FCA statement on AI and intermediation: “AI will relocate intermediation, not remove it, FCA says,” Mortgage Solutions, 6 July 2026.
FCA encouragement of broker AI adoption as part of mortgage market reforms, reported by Finextra and FStech, 2026.
UK financial services regulators’ approach to AI in 2026, including confirmation that the FCA will not introduce AI-specific rules, preferring principles-based oversight. Global Policy Watch, April 2026.
FINRA (Financial Industry Regulatory Authority) is the self-regulatory organisation that oversees broker-dealers in the United States under the supervision of the Securities and Exchange Commission.
Hassabis was awarded the 2024 Nobel Prize in Chemistry for his work on AlphaFold, which predicted the 3D structure of virtually all known proteins.
The Anthropic Mythos incident referenced in the Axios interview involved the Trump administration freezing Anthropic’s most powerful AI models via an export-control order in mid-2026, before negotiating their release over approximately two and a half weeks.