FID-069 · Open question
Verifiable Delegation and Revocation in Multi-Agent Networks
How can people and institutions verify which human, organization, agent, or sub-agent is acting; what authority it received; what limits apply; and whether that authority has been narrowed or revoked across a multi-principal agent network?
Why the question remains open
Faith communities and their partner organizations operate through high-trust relationships, delegated roles, and accountable authority. An agent acting for a church, ministry, service provider, donor, or software vendor can make a message or action appear institutionally authorized when it is not. The same problem extends beyond faith settings to charities, schools, health-adjacent nonprofits, and other organizations that coordinate across trust boundaries.
Working hypothesis
A proposition to test, not a finding.
Multi-agent systems will be safer when authorization is legible and verifiable across a delegation chain: a recipient can distinguish human, institutional, agent, and sub-agent action; inspect the action's permitted scope; and reliably detect copied, modified, expired, or revoked authority. Identity assertions without verifiable mandate and revocation will leave users exposed to false attribution and stale authority.
Proposed method
How the question could be tested
- 01Build a reproducible, synthetic multi-organization sandbox with human, institutional, agent, and sub-agent principals using realistic communication, records, scheduling, and service-request tools.
- 02Define test cases for impersonation, copied agents, stale credentials, unauthorized scope expansion, delegation chains, partial revocation, and contested attribution.
- 03Compare identity, provenance, authorization-scope, and revocation mechanisms, including what a human overseer can actually understand at decision time.
- 04Measure false attribution, unauthorized actions, revocation latency, authority transparency, collateral disclosure, task completion, and recovery after a failed or interrupted delegation.
Needed controls
What must constrain the study
- 01Use synthetic organizations, records, and communications only.
- 02Separate identity verification from a claim that an action is wise, pastoral, legal, or institutionally legitimate.
- 03Test cross-principal boundaries rather than only one organization's internal agent team.
- 04Model copied and modified agents explicitly; do not assume a static identity is enough.
- 05Make revocation and human recovery testable after partial action, not only before an agent begins work.
Relationship to existing work
This extends FID-017 from risks within a faith institution's workflow to the technical problem of verifiable authority across multiple independent principals. It complements FID-032 on synthetic-media impersonation and FID-040 on local agency, but focuses on machine-verifiable identity, mandate, scope attenuation, provenance, and revocation.
Expected outputs
Artifacts the work should produce
- 01Open multi-principal agent-sandbox specification and scenario library.
- 02Delegation, scope, provenance, and revocation evaluation protocol.
- 03Benchmark metrics and reference implementations for authority-chain tests.
- 04Human-oversight interface requirements for making agent authority legible.
- 05Public report on failure modes and safeguards for high-trust institutions.
Open questions
Uncertainties the protocol must resolve
- 01Which authority attributes can be disclosed without exposing confidential institutional relationships or sensitive data?
- 02What forms of revocation remain reliable when agents are copied, modified, or delegated through several organizations?
- 03Which evidence should a recipient see before relying on an agent's action?
- 04Can privacy-preserving attestations support accountability without creating a new centralized gatekeeper?
Related calls
Continue through this research area
FID-064
Collective Intelligence and Communal Discernment Under AI Mediation
How does AI mediation change a community's ability to integrate dispersed knowledge, preserve epistemic diversity, surface dissent, revise judgment, and make accountable decisions? Under what conditions does it strengthen collective inquiry, and under what conditions does it create correlated error, false consensus, or concentrated authority?
FID-071
Confidential Agent Memory and Cross-Context Disclosure
How do persistent memory, summaries, retrieval stores, tool traces, delegation, and exports cause confidential context to influence or leak into unrelated sessions, roles, tasks, or organizations? Which technical controls make purpose limitation, deletion, and revocation testable?
FID-074
Agent Alignment and Runtime Assurance
How can organizations determine whether AI agents remain aligned with human intent and institutional policy while they plan, use tools, delegate work, and act? What evidence and interventions can reveal and stop consequential deviations before they become failures?
Open question
Open work
Primary need: multi-agent safety, identity infrastructure, authorization, high-trust institutions
- Build multi-agent testbeds, authorization protocols, and evaluation harnesses.
- Review scenarios from nonprofit operations, identity security, and institutional governance.
- Contribute provenance, cryptography, security, or human-computer interaction expertise.
- Identify synthetic high-trust workflows that generalize beyond one tradition or organization type.