Determine feasible user control capabilities over AI agents

Determine the technological control mechanisms that end users can reliably employ to constrain and govern the actions of registered AI agents executing general, open-ended instructions, including capabilities for scope limitation, permissioning, and oversight.

Background

The authors discuss accountability regimes for AI agents, including potential requirements that agents be registered to identified humans. They highlight challenges stemming from increasingly general agent instructions and limits on user control, raising concerns for liability and agency law. Clarifying feasible technical controls is necessary to design effective accountability and governance regimes.

References

The trajectory of technological development is towards evermore general instructions (``go make \$1 million") and it is unclear what technological capacity users will have to reliably implement controls on what an agent can and cannot do.

— An Economy of AI Agents  (2509.01063 - Hadfield et al., 1 Sep 2025) in Institutions for AI agents, Subsection “Agent identity, registration, and records”

Customer governance is well covered, but whether this innovation truly constitutes transformation, rather than adaptation, remains to be seen.

— Assessing Company Contributions to Societal Resilience: Extending the Societal Capacity Assessment Framework to Agentic AI  (2608.27238 - Simons et al., 27 Aug 2026) in Appendix A, subsection “Microsoft Case: Findings,” subsubsection “Adaptive Capacity”