Asset Mgmt Agent Identity Identity | Moring
AI agent identity is the verifiable representation of an autonomous system: what it is, which human owns it, what authority it holds, and how that authority is issued, constrained, checked, and revoked. For asset managers, agent identity is the foundation every other control sits on, because unidentified agents cannot be governed at all.
Short answer
AI agent identity is the verifiable representation of an autonomous system: what it is, which human owns it, what authority it holds, and how that authority is issued, constrained, checked, and revoked. For asset managers, agent identity is the foundation every other control sits on, because unidentified agents cannot be governed at all.
How it works
Identity, for an AI agent, is a bundle rather than a login. Skyfire's buyer guidance defines it as four attributes: a verifiable identity, an accountable owner, bounded authority, and revocable access. Nensha Bardoliwalla's enterprise post makes the operational version blunt: as agents spread, most organizations cannot answer what every one of them is authorized to do right now, and the remedy he proposes issues every agent a verifiable identity at runtime covering what it can touch, what it may do, and who is accountable. That bundle is the asset management firm's actual unit of governance.
Why the bundle differs from a user account follows from agent behavior. The CSA Agent Identity Governance Framework describes agents as authenticating programmatically, acting at machine speed, spanning trust boundaries, spawning child agents, and persisting after the business context that justified their privileges has changed. Witness.ai adds the credential profile: ephemeral credentials rather than sessions, and delegation chains across sub-agents and APIs that legacy identity models were not designed to track. A borrowed account or shared service key hides every one of those properties.
The arithmetic explains the urgency. Witness.ai reports non-human identities, including agents, bots, and service accounts, already outnumber human users roughly three to one across enterprises and are frequently targeted in supply chain attacks. Lumos counted more than 450,000 agent actions in a single week in September 2026 at a company of fewer than 200 people, more than 2,000 actions per person. Identity is the only handle that survives that volume; a per-person review process cannot.
Identity quality determines what every downstream control can do. Approval gates only bind if the approving identity is distinct from the acting one, which is why Curity argues agents should stay machine identities and never blend with human identities. Per-call authorization only works if the credential presented reflects the agent's own scope rather than a platform default, which is why Curity also prefers asymmetric, key-bound credentials such as signed tokens and certificates over symmetric secrets and static API keys.
For an asset manager, the identity layer is where regulation attaches most naturally. SR 11-7 expects model risk governance commensurate with impact, and the impact of an agent runs through what its identity permits it to touch. SR 26-2 third-party guidance covers identity platforms operated by vendors. The EU AI Act imposes documentation and transparency duties on firms operating in the European Union, and a verifiable per-agent identity with retained issuance records is the artifact those duties describe.
Where identity programs most often fall short is the middle attribute: bounded authority. Verifiable identity without bounded authority produces excellent records of actions nobody intended, and Omada's guidance names the pattern, agents chaining decisions and delegating across systems while remaining inside existing permissions but beyond what anyone meant. Authority has to be scoped to the agent's registered purpose and re-verified continuously, not inherited from the team that built it.
The practical definition to carry forward: an asset management agent's identity is whatever an examiner could be shown as the complete answer to four questions. What is this agent, under whose accountability does it operate, what may it touch and do, and how is that access ended. If any of the four is unanswered, the firm does not have an identity program; it has an inventory with gaps, and the exposure lives in the gaps.
Frequently asked questions
What does identity mean when the entity is an AI agent rather than a person?
Identity for an agent is a bundle of verifiable attributes: what the agent is, which named human is accountable for it, what authority it holds over systems and data, and how that access is constrained and revoked, as Skyfire's buyer guidance summarizes. The CSA Agent Identity Governance Framework explains why this differs from a user account, since agents authenticate programmatically, act at machine speed, spawn child agents, and persist after the context that justified their privileges has changed. For an asset manager, that bundle is the governance unit.
Why should agent identities stay separate from human identities?
Separation preserves the integrity of every control that depends on identity. Curity's guidance argues agents should remain machine identities and never be blended with human ones, because mixing creates ambiguity that is hard to maintain, and it recommends asymmetric, key-bound credentials such as signed tokens and certificates over symmetric secrets and static API keys. If an agent inherits a person's account, approval gates cannot distinguish the approver from the actor, and an asset management firm loses the accountability separation its review processes assume.
What breaks when an asset manager relies on shared service accounts for agents?
A shared service key erases the four identity attributes at once: there is no verifiable per-agent identity, no accountable owner, no bounded scope attributable to one purpose, and no clean revocation path that ends one agent's access without ending everyone's. Witness.ai documents the scale at which this fails, non-human identities outnumbering humans roughly three to one and generating delegation chains legacy identity models cannot track, while Lumos counted more than 450,000 agent actions in one week at a small company, volumes no shared account can answer for.
How does agent identity connect to the asset management regulatory frame?
Identity is where regulatory obligations attach because impact runs through access. SR 11-7 expects model risk governance commensurate with a model's impact, and an agent's impact is defined largely by what its identity permits it to touch, while SR 26-2 third-party guidance covers vendor-operated identity platforms. The EU AI Act adds documentation and transparency duties for firms operating in the European Union, and a verifiable per-agent identity with retained issuance and revocation records is precisely the evidence those duties anticipate.