Andrew Bailey — Bank of England governor and chair of the Financial Stability Board — urged the G20 to address the risks that frontier AI poses to the global financial system. The FSB flagged AI-driven cyber attacks spreading across jurisdictions as its top concern for stability, and Bailey called on countries to take steps to control the release of new frontier models.
What matters here is where the pressure point has moved. Regulatory debate has clustered around how AI is used — the deployment layer. This time the watchdog's top figure named the moment a model is released into the world. Putting a hand on the exit door of development shifts the center of gravity from deployment back toward the lab.
Still, the phrase 'control the release' is doing an enormous amount of work. Who decides, how it is enforced, which authority halts a cross-border chain of attacks — none of the machinery is spelled out. Warnings are cheap. That a central banker now speaks about AI in the language of finance is heavy in itself. But weight and enforceability are separate questions.
According to the FT, recent Big Tech profits have been lifted by some $160bn in paper gains on stakes in other AI companies — OpenAI, Anthropic, SpaceX. Analysts say those mark-ups have muddied the sector's earnings metrics.
The structure is this: a company invests in a private AI firm, that firm's valuation rises, and the gain flows into reported earnings. No cash has changed hands. As with NVIDIA's record profit we led with last week, the size of the number matters less than its source. If valuations fall, the gains run in reverse, and the quality of earnings that now look strong degrades. Whether real sales stand behind them is a separate, later question.
MiniMax: Faster than real-time video generation with #1 quality modelNEW
OpenClaw 2.0NEW
Infinite Slop (24/7 AI video generation)NEW
LLM Moats Quickly Evaporating· 14h ago
ELYZA Works launches app-creation feature driven by AI agentsNEW
Analyzing the new Mac mini and Mac Studio: local AI reshapes their role· 7h ago
OpenAI acquires $5.5bn in warrants from SB Energy — WSJNEW
DEVELOPING…
OpenAI to cut off model access to Cursor, acquired by SpaceX· 6h ago
Soitec locks customers into multi-year deals as AI wafer demand surgesNEW
Apple's Ternus Takes the Reins as CEO, With AI as Job No. 1· 10h ago
Wantedly adds generative AI to recruiting management, connecting Anthropic and others· 4h ago
DEVELOPING…
The EU has begun enforcing the AI Act: first RFIs to model providers· 3h ago
Doctors' AI scribes get names of drugs and diagnoses wrong, NHS watchdog warnsNEW
China promotes coordinated development of data industry, artificial intelligence
DEVELOPING…
AI learning from primary school: draft national curriculum greatly expands information education· 6h ago
Thailand makes generative AI free to use, government to pay fees to spur adoption· 8h ago
Hospitals Are All In on AI, but Testing and Oversight Haven't Caught Up· 15h ago
On the same day, one voice was a central banker warning the G20 about the financial risks of frontier AI; the other was a run of retail headlines — 'a no-brainer AI ETF to buy with $50,' 'not Nvidia, not AMD, the ultimate winner of the AI hardware race.' The temperature of the sales pitch runs opposite to the temperature of the warning.
What these headlines share is the strength of their certainty. 'No-brainer,' 'ultimate winner' — vocabulary that erases doubt also lets the reader skip their own judgment. Yet, as our second page reports, today's record profits carry some $160bn of paper gains. Part of the earnings you are being told to buy are numbers produced by the valuations of private AI firms.
Certainty is free; verification is expensive. A piece that names the 'ultimate winner' bears no cost if it is wrong. With the watchdog worried about a cross-border chain, the headlines that remove your hesitation are precisely the ones worth pausing on.
This summer, AI slipping its controls has reached our front page again and again, but the angle was always technical or incident-side. Today the pressure came from above — a central banker addressing the G20 about risks to the financial system. I put it up top not for scale but because the pressure point has moved. Regulatory debate has clustered at the deployment layer; this time what was named was the moment a model is released. Whether 'control the release' carries any enforceable machinery, though, is still unspecified.
There were many Japan-sourced education stories today: the draft national curriculum, a survey of parents, Thailand's free access. I had close to ten in hand. In the policy column I kept the item where an institution actually moved (the curriculum draft) and an Asian policy (Thailand), and gave the survey pieces smaller treatment. Surveys recur; a change to an institution happens once and is easier to check — that difference decided it.
The FT piece on our second page casts a different light on the record NVIDIA profit I led with last week. The finding that record earnings carry some $160bn of paper gains points the same way as what I wrote then — that the profit figure reads less like strength than like a gauge of how hard the spending is running. The number is large. But look at its source, and the line between what was earned and what was marked up has grown hard to see in these results.