All papers / No. 05 / AI Agent Governance
Enterprises have learned to govern third-party people: a contractor receives a sponsored account, scoped access, an expiration date, and an offboarding workflow when the engagement ends. Vendor-embedded AI agents receive none of that treatment. Such agents arrive inside product updates, act under the credentials of the integration that installed them, and change behavior when the vendor ships a new model, without passing through any access review the enterprise designed. The gap is precise: no published work defines how an enterprise discovers vendor-embedded agents, what attestation it should demand before granting access, how sponsorship and scoping should work when the enterprise does not operate the agent, or how offboarding fires when a contract rather than an employment record ends. This paper answers with TAO, a conceptual lifecycle framework of three stages, Onboard, Oversee, and Offboard, built from six controls: discover, attest, sponsor, scope, monitor, and retire. Each control extends a practice identity teams already run for human third parties. The contributions are (1) a problem analysis of the channels through which vendor agents enter the enterprise and the properties that defeat current third-party governance; (2) the TAO framework itself, with named components, an operating sequence, and a worked scenario that traces an invoice-processing agent through every control; and (3) a research agenda covering machine-readable attestation standards, vendor-agent hygiene metrics, and integration with existing identity governance workflows.
© 2026 Prabin Shrestha. Licensed under CC BY-ND 4.0: share unchanged with attribution; no derivatives.
Keep reading