Topology is where agency actually lives.
Agency is not a property of the agent in isolation. It's a property of the agent's position: its dependencies, who coordinates it, what authority flows through its edges. Rewire the topology, change the agency. This is as true of an org chart as of a multi-agent system.
The framework makes that position explicit.
@AnthropicAI's actual problem was a binary: serve everyone (government objects) or serve no one (customers lose).
Idea: A model-plus-provable-control-plane architecture offers a third option: an AI company can provide a model to a foreign player because that player's verifiable control plane provably bounds what the model can do: the cybersecurity-uplift risk that triggered the order is constrained by construction. The control plane is the export-compliance artifact. It's what makes "providing the model" defensible.
This means, tomorrow, the provider and the regulator can inspect the deployer's control plane, as a formal, auditable specification of what the model is permitted to do.
The AI sovereignty debate is stuck on the model: who trains it, where the weights live. Another real sovereignty question is the control plane: who decides what the agent is allowed to do. You can run a domestic model on foreign control logic and have surrendered the thing that mattered.
The safest architecture for an agentic economy is also the most sovereign one: the model proposes, a deterministic policy layer disposes. If the proposing model is foreign but the disposing layer is yours, you keep control.
New @iei_org compte rendu of the recent closed-door roundtable on the EU's DeFi consultation organized by @nathalie_aryana. The real questions are about the perimeter: circumvention, gatekeeping, regulated entities touching open infrastructure.
It also covers our proposal for a taxonomy of function and control, and a property catalogue that makes assurance verifiable.
https://t.co/ibCNCmA8RA
New note by the Intelligence Economy Institute.
A structured read of the "Digital Assets: the Future of Finance?" roundtable from the Paris Finance Forum this morning.
Adam Baker (@BlackRock), @ddisparte (@circle), Valérie Urbain (@EuroclearGroup), Mandy DeFilippo (@StanChart), Jean-Jacques Barberis (@CACEIS), moderated by @ClaudiaECohen (@business).
The short version: the architecture decisions of the next one to two years will define the next decade, and we should actively build the trust layer it runs on.
Full overview and analysis here: https://t.co/FD0vVg5Y0o