Trust for AI Agents — An Ecosystem-LevelFramework
This concept paper addresses ecosystem-level trust for AI agents: what evidence entities need in order to trust each other, and how it can be produced and conveyed. We argue that, as with people, trust is a risk decision on an interaction, based on a variety of information collected in several steps over time. Some of this information is associated with the entities, some with the content and context of the interaction. Unlike people and organisations, digital entities are poor accumulators of trust (ephemeral, mutable, duplicable). Trust in digital entities therefore relies more on the durable entities connected to them (principal, operator, execution environment) and on the content and context of the interaction itself. We also discuss how far this evidence can itself be trusted, and outline directions for further work.
Authors
- Luca Boldrin (ORCID: https://orcid.org/0000-0001-8403-1506)
- leone riello
- bo fjelkner
- ward duchamps
Institutions
- Ericsson (Sweden) (SE)
- University of Padua (IT)
- Thales (Belgium) (BE)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-19
- DOI
- https://doi.org/10.5281/zenodo.24361360
- Primary Topic
- Access Control and Trust
- Type
- article
- Field-Weighted Citation Impact
- 0.00