Greetings to all Free Lunch readers — I hope you have had a rewarding (northern hemisphere) summer. Your holiday may have more profound importance than you realise: to seek novelty and discovery is a fundamentally humanist act. Many thanks to all my wonderful colleagues who have stepped in for both Tej and myself to keep your favourite global economy newsletter going through August. Central bank governors have ended their holidays, with the annual Jackson Hole symposium and a US-hosted G20 meeting with finance ministers both taking place in the past week. Beyond the immediate question of where on earth US monetary policy is headed, policymakers are being urged to pay attention to some deep changes in our economies wrought by AI.

Andrew Bailey, who not only governs the Bank of England but chairs the international Financial Stability Board of the world's top financial policymakers, warned finance ministers and central bankers joining the G20 meeting this week of significant risks posed by AI to the financial system. Note the plural — not one, but two dangers featured in Bailey's letter. It's as though, if AI does not kill you one way, it will come back to do so in another. The first and more recognisable risk is that of a fall in asset values. Bailey doesn't use the word, but this is AI as the latest financial bubble. He warns against rises in leverage — piling large valuations on small amounts of equity — which is historically what blows the bubble bigger when things are good and worsens the crash afterwards. The second risk is superficially familiar too: Bailey warns that AI could intensify the cyber risks threatening to disrupt institutions and markets. But there is a deeper danger below the surface. Listen to Bailey: The biggest issue here, I think, isn't simply that some functioning could be disrupted. It is that when we can no longer trust that services we have come to rely on will always be available, the system works less well even before any disruption occurs.

At Jackson Hole last week, Princeton economist Markus Brunnermeier argued that agentic AI presents entirely novel challenges to how economies work. New concepts are needed for economic analysis, which may overturn the established conclusions of conventional theory. Brunnermeier's own conceptual innovation is "asymmetric understanding", which deepens the old information economics concept of asymmetric information. Asymmetric information is when someone knows more about someone else than the other knows about them (or about matters of relevance to both). But this imperfect knowledge still operates within a correct understanding of how the world works (so that if you had perfect information, you would predict things correctly). Brunnermeier considers cases where not just information but understanding is imperfect and indeed asymmetric, and argues that in a world with AI agents, this asymmetry is to humans' disadvantage: we cannot understand how AI agents think, even as they understand (at least to the point of being able to predict our actions better than we do theirs) how we think.