2026-06-17-aria-ccti-proposal-draft-2

1. SCHEDULE 3

2. PART 1 - SERVICES

   1. Project Name

   Centre for Cryptographic Trust Infrastructure (CCTI)

   2. Background

   [Guidance: set out here any relevant background or context to the Project]

   3. Aims and objectives of the Services

   AI agents can find counterparties, negotiate terms, and execute real-world transactions without the prior trust history that humans depend on, provided that the agents’ claims about their physical entities and their reasoning are verifiable. This project’s north star is to build the necessary infrastructure for such proofs and to test this hypothesis empirically on at least one vertically integrated use case. By implementing this cryptographic trust infrastructure and testing it with AI agents that initially do not trust each other, the project team aims to demonstrate that this approach allows AI agents to come to an agreement where humans or AI without such a trust infrastructure would not.

   In addition, the project team will serve as the cyber-physical trust-layer facilitator. First, toward the other creators, it will build and maintain reusable trust primitives that they can adopt and build upon, and convene the programme around their convergence. Second, toward ARIA’s Arena, it will contribute its cyber-physical trust expertise to the design of benchmarks¹ and challenges that to exercise the programme’s trust primitives. The guiding leitmotif of the facilitation work is: CCTI commits to the offer and to the contribution effort; adoption decisions remain with the creators as well as with ARIA and Arena leadership to steer.

   Through this work, the project team aims to be a centre of excellence at the intersection of information security, cyber-physical systems, and agentic AI in the UK.

   4. Services

      1. Funded Activities

   Outputs:

3. A software system that can collect and aggregate cryptographically attested evidence of the physical world from many sources, derive from it verifiable claims in zero knowledge for particular use cases, and make these available to AI agents for autonomous interaction (WP-A);

4. A zero-knowledge database (zkDB) system and query language that enables domain specialists without cryptographic expertise to define systems similar to WP-A and WP-C for new use cases (WP-B);

5. An implementation of the Epistemic Integrity Certificate (EIC) and credible matching – a novel approach that allows agents to (i) privately commit their beliefs and conditions for cooperation with other agents, (ii) perform secure match-making over their private beliefs, and (iii) to negotiate the (further) evidence needed to come to an agreement (WP-C);

6. A demonstration of the EIC and trust infrastructure on at least one vertically integrated use case (ideally two contrasting use cases), empirically evaluating the ability of AI-with-CCTI to come to an agreement as compared to AI-without-CCTI and human negotiation (WP-D);

7. Conditionally on the success of the previous outputs, an analysis of the new institutional cooperation patterns that emerge in AI-with-CCTI collaboration (WP-E);

8. To structure the integration and facilitation role, (i) a (living) document of the programme cohort and Arena roadmap covering dependencies, gaps, synergies among the cohort’s creators, (ii) a convergence method (a cadence and process for iterating toward candidate shared primitives with willing teams), and (iii) a contribution to ARIA’s Arena: one or more open-source benchmarks for trust-primitive verification, plus proposed challenge designs (WP-F).
   All artefacts – publications, research outputs, code, integration patterns – will be open access and open source by default, with the exception of any proprietary data or tools provided by industry partners.
   Architecture and Work Package Structure
   The project’s technology stack consists of the following layers:

• At the lowest layer, the project team will use existing mechanisms for verifiable and privacy-preserving computation: TEEs, zk-SNARKs, MPC, and zkVMs. The team aims to eventually use formally verified implementations where possible, but will pragmatically get started with whatever currently exists. TEEs will be the fastest way of getting started, but the team aims to eventually minimise its dependency on trusted hardware except as part of a defence-in-depth strategy.
• On top of this foundation the project team will develop support for several cryptographically attested types of evidence and their provenance, including sensor data (claims about the world), business data (claims about cross-organisational interaction), EICs (epistemic claims about an agent’s reasoning), and recursive proofs (claims supported by another instance of this project’s trust infrastructure).
• To process this evidence the project team will initially develop use-case-specific logic in the form of TEE enclaves, SNARK circuits, and/or zkVM programs. As the requirements for evidence processing become clear, the team plans to generalise the common elements from different use cases into a scalable rule system inspired by database query engines, and develop a language for expressing the use-case-specific logic in a concise, declarative way.
• The top layer is the application layer. Here the project team will build upon the lower layers to implement matching and reputation. The team will also prototype LLM-based agents to experimentally test their ability to make use of the trust infrastructure, and to measure the impact of this infrastructure on their ability to come to an agreement, compared with humans and AI agents without the trust infrastructure.
  CCTI runs without formal phases; all work packages execute in parallel from the start of the project (with the exception of WP-E, which is conditional on the success of the other work packages).

WP(Lead)	Title	Activity Period	Primary Deliverables
WP-A(Cambridge)	Attested-data layer + Attestation Types	Jul 2026 to Dec 2027	Open-source reference implementation for heterogeneous attestation types and ZK-provable multi-dimensional reputation
WP-B(Cambridge)	Verifiable query/rule layer (zkDB)	Jul 2026 to Dec 2027	Verifiable querying over heterogeneous attested facts; domain-rule expression, reducible to general-purpose code
WP-C(SINE)	EIC + credible matching	Jul 2026 to Dec 2027	EIC definition and reference implementation over conjugate-family beliefs; credible matching protocol implementation
WP-D(SINE / Cambridge)	Integration showcase + evaluation of AI advantage	Jul 2026 to Dec 2027	At least one vertically integrated EIC-bearing use case (aspiration: two, the second contrasting with the first), end-to-end, with comparison of AI-with-CCTI to AI-without-CCTI and human
WP-E(SINE)	Institutional lens	Conditional, from Jul 2027 to Dec 2027	Characterisation of one institutional equivalent (e.g. insurance, arbitration, or reputation) emerging among agents in the AI-with-CCTI collaboration setting
WP-F (SINE / Cambridge)	Integration Facilitation	Jul 2026 onwards; scaling conditional on demand	Supporting documents for the facilitation role: an integration map, a joint convergence method, contributions to ARIA’s Arena (benchmark(s), challenge design proposal(s))

How Work Packages, Deliverables, and Milestones Fit Together

The work packages are interdependent rather than sequential: WP-A produces verifiable attestations; WP-B makes them queryable; WP-C supplies novel primitives, such as the EIC, and an application on top of the EIC (credible matching); WP-D wires both into a real-world use case and measures the resulting agent behaviour; and WP-E, if activated, characterises the institutional structures that emerge. WP-F covers the foundations for successful facilitation, covering knowledge products and a convergence method for programme cohort-wide coordination.

The Go/No-Go Milestones correspond to key risks where the interdependence of work packages is most fragile: (i) whether the attestation layer integrates and a demonstrator exhibits a cooperation-unlocking decision (M2); (ii) whether the query engine provides enough shared functionality across two use cases to justify keeping it as a language, or whether it should be downgraded to general-purpose code (M4); (iii) whether the integrated stack produces a measurable AI-Advantage effect and emergent behaviour warranting institutional analysis (M4); (iv) whether that institutional structure is legible enough to characterise (M5).

Open questions. The evaluation use case(s) are not yet selected (the project team will do this by M1). Whether the planned approach will unlock cooperation in these use cases is unknown, and will be investigated during the first year. The team will attempt to obtain real industry data for the use cases, but this is not required – anonymised or synthetic data is an acceptable fallback.

Our Role as a Facilitator

Beyond its own research, CCTI acts as the programme’s cyber-physical trust-layer facilitator.
First, towards other creators, it builds and maintains reusable trust primitives (attested-evidence and EIC), infrastructure (such as a potential trust language) and convenes the programme around them. Second, toward ARIA’s Arena, it contributes its cyber-physical trust expertise to the design of benchmarks and challenges that test the production and verification of trustworthy evidence.
Given the CCTI’s role, it commits to the offer and the contribution effort ,while the eventual uptake hinges on creators and on incorporation into the Arena. Two programme-facing checkpoints instrument this role: an early integration reconnaissance that maps cohort and Arena dependencies and course-corrects the work plan once the cohort is known (M1), and a mid-point integration review that reports separately on what CCTI offered versus what others adopted, and conditionally activates a support track if uptake warrants it (M4).

Open questions. The evaluation use case(s) are not yet selected (the project team will do this by M1). Whether the planned approach will unlock cooperation in these use cases is unknown, and will be investigated during the first year. The team will attempt to obtain real industry data for the use cases, but this is not required – anonymised or synthetic data is an acceptable fallback.

2. The following Milestones will be used to monitor your delivery of the Project. The costs associated with completing these Milestones are estimated in the cost breakdown sheet [INSERT Link]:

ID	Type	Milestones	Measure	Date	Notes
1	Milestone	Setup, Use Case Selection, and Specification; Integration Mapping	Team onboarded and open positions recruited; shortlist of EIC use cases selected; design docs for WP-A, WP-B, and WP-C; first version of the integration map delivered (dependencies, gaps, synergies) + convergence method proposal	30/9/2026
2	Go/No Go Milestone	Attestation-Layer Feasibility	Prototype implementation of 2 or more attestation types, and the EIC; proof-of-concept query language design and TEE-based database implementation; end-to-end demonstrator covering WPs A to C	31/12/2026	First gate; on Go, reusable attestation components may start early; No-Go simplifies integration scope
3	Milestone	Depth and Two EIC-based Implementations	WP-A handles ≥3 attestation types; WP-B/WP-C matching end-to-end with multi-dimensional reputation; at least one EIC-based use case implemented as a demonstrator; a demonstrator exhibiting a cooperation-unlocking decision	31/3/2027
4	Go/No Go Milestone	Language Gate + AI-Advantage Signal & WP-E Commit	Language gate: query language has demonstrable advantages over general-purpose code across the two use cases (No-Go = replace with general-purpose code). AI-Advantage gate: comparison of AI-with-CCTI, AI-without-CCTI and human coordination on industry use cases with effect sizes and confidence intervals; publishable either way; WP-E launch decision based on observed patterns.Integration Review (WP-F): offered-vs-adopted ledger: a) CCTI’s primitives versus uptake in the programme and at Arena(s), b) outcome of proposed challenge designs; benchmark(s) released	30/6/2027	No-Go on AI Advantage triggers a headline-claim pivot
5	Go/No Go Milestone only if WP-E is activated	Scale-Up and Institutional Probe	WP-A attestations and WP-B query engine use cryptographic security without assuming trusted hardware; larger-scale AI advantage evaluation; preliminary WP-E observations if active	30/9/2027	Reverts to a Project Milestone if WP-E was not activated
6	Milestone	Final Publication and Handover	Submitted papers on zkDB language and implementation, AI advantage evaluation, Coasean qualitative-framing analysis, and (if WP-E = Go) institutional lens; final showcase release with integration-pattern documentation; CCTI continuity plan submitted	31/12/2027	Project closeout; defines follow-on agenda

In addition to the above Milestones, the following project plan will be used to track progress throughout the project and may be updated from time to time: [insert link to project plan/gantt chart].

3. The intended long-term outcomes are summarised below:

   1. Continuity of CCTI through industry partnerships, follow-on research funding, or the Cambridge-anchored community hub, sustaining the consortium as a long-term node in the UK’s cyber-physical security landscape.

      2. Open source trust infrastructure centered around a general-purpose zero-knowledge database (zkDB) with support for various types of cryptographic attestation, a declarative language for domain rules, and example instantiations on real-world use cases.

         3. The EIC as a reusable primitive: a credible commitment to epistemic state that other Scaling Trust creators and ARIA Arena challenges can adopt, directly instantiating the programme’s question of when AI-specific primitives provably expand achievable coordination outcomes.

         4. Adoption pathways for the attested-data and query layers in supply chains, manufacturing, logistics, and adjacent domains where coordination without prior trust history is currently expensive or blocked.

         5. Methodological precedent for empirical AI-Advantage measurement, transferable to other creators and to Scaling Trust Arena challenges.

         6. Theoretical foundation for follow-on work on agent-mediated institutional dynamics (insurance, arbitration, reputation), should WP-E produce promising preliminary results.

   5. Principal investigator/key researchers

      1. For the purposes of clause 22.7, the following principal investigators and key researchers have been identified:

Name	Role	Organisation
Martin Kleppmann	Co-PI	University of Cambridge
Martin Pompéry	Co-PI	SINE Foundation

6. Sub-Contractors

   1. For the purposes of clause 19, the following Sub-Contractors have been agreed:

Name	Description of activity	Location of organisation
SINE Foundation e.V.	Lead for WP-C and WP-E; co-lead for WP-D and WP-F; expertise in cryptography, game theory, and industrial supply chains	Germany
Ink & Switch UK Ltd.	Expertise in zk-SNARKs, zkVMs, software engineering, and human-agent interaction	UK
Light Squares Ltd.	Expertise in TEEs and cryptographic protocol engineering	UK

Footnotes

1. Attested evidence formats and infrastructure, Epistemic Integrity Certificate (EIC), as well as coordination primitives as described below ↩