REASSURE: Real-world Evaluation and Assurance for Safe and Trustworthy Responsible AI Ecosystems

Summary

Frontline services, financial services, and other public-facing sectors are adopting AI rapidly, often without mature assurance processes that are adequate for complex real-world and sociotechnical settings. This campaign develops and tests responsible AI assurance approaches across three linked workstreams: FRONTLINE, which focuses on universal support and public-interest services across education, health and the third sector; PROTRACE, which develops protocols and traceability for agentic AI in financial services; and Methodologies and AI Infrastructures, which provides the shared sociotechnical methods, participatory approaches, and AI tools needed to coordinate learning across use cases. Together, the campaign will deliver sociotechnical assurance methods, demonstrators, governance tools, and evidence for safer, more trustworthy AI deployment.

The overall aim of the campaign is to develop, implement, and evaluate practical sociotechnical assurance approaches for AI systems deployed in real-world, high-impact settings, as informed by a use-case specific, exploratory approach. The required use cases will come from a focus on frontline support services, agentic AI in financial services, and the reusable methods and infrastructures needed to coordinate learning across assurance contexts. Objectives include:

• Map the sociotechnical challenges, risks, harms, and governance gaps associated with AI adoption across frontline support services and agentic financial-service settings.
• Co-produce assurance requirements that inform AI development, procurement, deployment, and monitoring, and that align with service needs, user and provider concerns, regulatory expectations, and technical processes.
• Develop and validate lifecycle-oriented assurance approaches that connect PRISM, AdSoLve, and PHAWM approaches with participatory and comparative methods.
• Create multi-agent protocols for dependable interaction among AI agents in financial services.
• Develop and demonstrate explainable traceability techniques that support compliance with regulatory obligations such as the FCA Consumer Duty.
• Collate and compare methods, tools, frameworks, and AI infrastructures used across the campaign and the AI Assurance landscape to produce a reusable assurance methodology, a catalogue of approaches, and cross-case insights for the emerging AI assurance ecosystem.