Originally published by ARIA on 17 November 2025, and updated to v2.0 on 10 February 2026 (see below). Read the latest version (DOI record) or the original v1.0.

Version 1 of our programme thesis is out, and we want your feedback!

AI is collapsing coordination costs, signalling a new era for our society and our economy. The Scaling Trust thesis envisions this era as one of many AI agents acting as our faithful representatives, able to go out into both digital and physical worlds and mobilise, negotiate, and verify on our behalf.

The proposed programme suggests three mutually reinforcing efforts to reach this vision: advancing fundamental research into agentic coordination; building open-source tools to make this capability robust and available to all; and setting up a series of challenges that will test AI systems’ capabilities in real environments, with multi-million pound prize funds for the best teams.

Read the thesis and share your thoughts.

The thesis, simply stated

Every day we spend time, effort, and resources coordinating with each other – finding the right people or services to work with, negotiating the terms of the interaction, and enforcing the agreements made. These are fundamental costs of modern life we pay, what economists call ‘Coasean transaction costs’, and AI is collapsing them. Their reduction signals a new era for our society and our economy.

We envision this era as one where AI agents are our faithful representatives, continuously aligning to our preferences, able to go out into both digital and physical worlds and cheaply mobilise, negotiate, and verify on our behalf. As opposed to an all-powerful, all-knowing AI, this is a world of many-agents, each holding our private information, operating with their own objectives, and constantly interacting with institutions, humans and other agents.

This is an exciting vision, one of human flourishing and augmentation via technology, that preserves our plurality and uniqueness.

We believe new security primitives like programmable cryptography and secure hardware represent a unique opportunity to usher in this new world by:

1. Creating a scalable trust infrastructure for agents across digital and physical worlds, thereby increasing the complexity of secure interactions agents can engage in, and the pool of potential parties they can engage in them with;
2. Enabling new forms of secure interactions previously impossible for humans or traditional software, unlocking new valuable markets and societal value.

To this end, we plan on launching three core initiatives. First, we will fund open-source applied tools needed to build this capability robustly, safely and for all of humanity. Second, we will fund fundamental research to build a stronger theoretical foundation for this field, and to uncover new security primitives agents can harness for secure coordination. Third, we will launch a series of challenges open to all to benchmark the tools built and research developed, with multi-million-pound prize funds for the best teams.

It will take many disciplines and stakeholders to make this vision a reality. The road ahead is riddled with risks, but getting it right is transformative. We’re calling on all of you – hackers, roboticists, game theorists, cryptographers, AI security engineers, and everything in between – challenge our plans, join the community, and help us Scale Trust.

The thesis was shaped with the help and feedback of many, including Nora Ammann, David ‘davidad’ Dalrymple, Orr Paradise, Eder Medina, Edith-Clare Hall, Sarath Murugan, Harry Jenkins, Quintus Kilbourn, Logan Graham, Daniel Freeman, Evan Miyazono, Lewis Hammond and all Discovery Workshop participants. Thank you!

Update, 10 February 2026: v2.0

We released v2.0 of the thesis alongside our first call for proposals. From its update log:

As we release our first solicitation, we want to update our thesis to share our latest world view with applicants and the broader community. This v2.0 fundamentally stays the same in its broad strokes as v1.0; however, there are elements that have changed as we continue to learn from conversations and iterate our thinking.

What has changed:

• Added ‘Contracts’ mental model with example applications to help concretise the outputs of the programme.
• Added Programme Objectives.
• Updated details on sub-components.
• Updated details on the Arena.
• Refined research agenda, in particular added ‘Foundations of Generative Security’ and removed ‘AI Game Theory’ as a core pillar.
• Deleted ‘What we do not expect to fund’ as it will implicitly live in Solicitation.
• Updated ‘What we are still trying to figure out’ to reflect our latest questions.

The summary above is otherwise unchanged, except that the “series of challenges” is now the Arena. It is still a work in progress: if you spot a mistake or want to push back on something, tell us, and help us get to v3.0.