A Princeton economist has told the world's most powerful central bankers that artificial intelligence might soon understand their own policies better than they do — and his proposed remedy is strikingly unconventional: hold two separate press conferences, one for humans and one for machines.

Markus K. Brunnermeier presented his paper on August 29 at the 2026 Jackson Hole Economic Policy Symposium, the annual gathering in Jackson Hole, Wyoming, where central bankers, academics, and policymakers debate the biggest questions facing the global economy. Papers delivered there have historically shaped monetary policy debates well beyond the conference room. This year, AI dominated the conversation. Brunnermeier's contribution, however, focused less on productivity gains and more on an uncomfortable possibility: that the tools might soon be smarter than the toolmakers.

The asymmetric understanding problem

At the core of Brunnermeier's argument is a concept he calls "asymmetric understanding." Traditional economics has long dealt with information asymmetry, in which one party to a transaction knows more than another — for example, a used car dealer who knows the engine is failing while the buyer does not.

Brunnermeier's version reverses the dynamic. In his framework, AI agents could develop knowledge of monetary policy that humans, including the policymakers themselves, cannot fully interpret. The machines would not merely have more data; they would possess a fundamentally different — and potentially superior — grasp of how policy decisions ripple through the economy. The concern is not hypothetical in kind: algorithmic and machine-learning systems already play a substantial role in trading and market analysis, and any AI system capable of systematically anticipating central bank reactions would change how markets price policy news.

Speak unpredictably, or get gamed

Brunnermeier's proposed countermeasures resemble tactics for outmaneuvering a mind-reading opponent at poker. He suggested that central banks may need to adopt a more opaque and unpredictable approach to their communications, effectively reversing years of movement toward clarity and openness — a shift with real stakes, since central banks, notably the U.S. Federal Reserve, have spent decades building forward guidance and press-conference transparency specifically to stabilize expectations and reduce market volatility.

The most striking recommendation was the dual press conference concept: separate briefings for human audiences and machine audiences, each calibrated to the way those audiences process information.

He also called for robust regulatory frameworks that simplify oversight, with the aim of preserving what he described as market confidence and informational integrity.

The presentation stood in notable contrast to the views of Fed Chair Kevin Warsh, whose preceding keynote took a more optimistic tone on AI as a productivity-enhancing agent. Warsh highlighted the growing investment landscape around AI, noting that AI token sales have exceeded $100 billion annually, an increase of over 500% year-over-year as of August 2026.

What this means going forward

Attendees at the symposium reportedly found the presentation useful for framing long-term AI risks, even as other policymakers such as Boston Fed President Susan Collins emphasized more immediate applications.

The symposium, held August 27 through 29 in Jackson Hole, Wyoming, examined AI's economic potential across multiple sessions. But it was Brunnermeier's closing paper that left the room with the hardest question: what happens when the audience for monetary policy is no longer human — and the humans are the ones who can't keep up?