Informatics team selected for ARIA programme developing trusted AI infrastructure

A team of researchers from the School of Informatics has been selected as part of the first cohort funded through the Advanced Research + Invention Agency (ARIA) Scaling Trust programme, a major UK initiative supporting the development of secure and trustworthy AI systems. The team is led by Tariq Elahi and includes Myrto Arapinis, Jianyi Cheng, Michele Ciampi and Wenda Li.

Building secure foundations for AI agents

The project, Agentic Security Reasoner for Emerging Multi-agent Protocols, has been selected within the programme's tooling track and aims to develop infrastructure that allows AI agents to negotiate, verify and enforce security and privacy requirements automatically.

The project is one of the first selected through ARIA's Scaling Trust programme, which is supporting research into the infrastructure needed for AI agents to collaborate safely and reliably across digital and physical environments.

ARIA Scaling Trust

Combining AI, cryptography and formal verification

The project combines expertise in cryptography, formal verification, AI reasoning and secure systems to help address one of the key challenges facing future AI systems: how autonomous agents can coordinate and collaborate safely on behalf of people and organisations.

The team is developing an agentic security stack that can:

  • Interpret security and privacy requirements
  • Translate tasks into formal specifications
  • Select or generate appropriate cryptographic protocols
  • Verify security properties before actions are taken
  • Enable trustworthy interactions between multiple AI agents
We want AI agents to work out for themselves how to keep a task secure and private, and to prove they've got it right before any actions are taken. If we succeed, assurance will be built into every interaction, from the lead agent to its helpers and to other agents.
Tariq Elahi, project lead Assistant Professor, School of Informatics, The University of Edinburgh

Part of the ARIA Scaling Trust programme

The project forms part of ARIA's wider Scaling Trust programme, which is funding open-source tools and fundamental research to support secure AI coordination across digital and physical environments.

Everything developed through the programme is intended to be open-source by default, helping ensure that future AI infrastructure remains transparent, contestable and accessible for others to build upon.

Looking ahead

The selected projects will contribute towards the development of the Scaling Trust Arena, a UK-based testbed where AI systems will be evaluated under realistic and adversarial conditions.

The work positions the University of Edinburgh at the forefront of efforts to ensure that future AI agents can interact securely, protect privacy and provide verifiable assurances about their behaviour.

Read more

This article was published on