AgentCustody

Agencies increasingly operate automation and agent workflows on behalf of several clients. A client needs to know which agent acted, under whose authority, with which credential and whether the external system actually changed. AgentCustody proposes per-client identities, capability grants, action receipts, scope-violation detection and a shareable monthly evidence report. The supplied research confirms active enterprise agent-identity investment, an emerging identity standard effort and open capability-token infrastructure, while finding no agency-facing product focused on client-separated audit trails.

Cryptographic identity does not prove that an action was authorized, correct or beneficial. Agent registration, software version, client tenant, human principal, capability grant, credential reference, intent, approval, tool command, provider acknowledgement, destination readback, policy result, exception, revocation and business outcome must remain separate. Secrets should never enter the report or public transparency log.

Tamper evidence proves a bounded record has not changed after signing; it does not prove the event payload was true or complete. An agency cannot use the report to replace client review, security assurance or contractual accountability. Per-client history may create switching cost, but export and verification must remain possible so the trust artifact does not become captivity.

An automation agency, managed service or professional-services firm operating agent workflows under distinct client credentials and approval boundaries.