Summary

Artificial intelligence is increasingly being deployed as autonomous “agents” that act on our behalf in homes, workplaces, and public services, such as healthcare where they can monitor wellbeing, raise alarms, and delegate tasks to other agents. While such capabilities offer significant benefits, they also create an urgent accountability challenge: when one AI agent delegates authority to another and something goes wrong, who is responsible, and how could this be proven?

Project "Agent for Good" addresses this by developing a “Digital Agent Passport” for AI agents: a cryptographically secured record of an agent’s identity, permissions, and the authority that granted them. As agents act and interact, their activities and delegations are recorded in a traceable, verifiable, tamper evident Chain-of-Accountability. This creates a reliable audit trail that can identify where decisions were made and who should be held accountable when failures occur.

We will develop the technical foundations and governance guidelines of this approach in a simulated smart-home care-monitoring environment, deliberately introducing failures to validate the effectiveness of the accountability chain. We will also engage stakeholders from industry, academia, and public bodies to define meaningful accountability for agentic AI and inform practical guidance for deploying trustworthy, auditable and safe agentic AI systems.