Book a demo
Topics · GOVERNANCE-AGENTS

Asset Mgmt Agent Identity Quarterly Review | Moring

A quarterly review of AI agent identity checks every agent's scope against its registered purpose, verifies delegation chains and drift remediations, and retires access that no longer matches. At an asset management firm the cadence is a governance commitment: agents accumulate entitlements faster than humans, and the review is what keeps least privilege true.

On this page
  1. Short answer
  2. How it works
  3. FAQs
  4. Related reading
  5. Sources

Short answer

A quarterly review of AI agent identity checks every agent's scope against its registered purpose, verifies delegation chains and drift remediations, and retires access that no longer matches. At an asset management firm the cadence is a governance commitment: agents accumulate entitlements faster than humans, and the review is what keeps least privilege true.

How it works

The quarterly review exists because agent access decays between reviews. Orchid's guidance covers lifecycle automation and continuous verification as the design response, but the review is the governance checkpoint where drift findings become decisions: scopes that no longer match purpose are narrowed, exceptions are resolved with recorded rationale, and decommissioned agents have their access history archived. An asset management firm that reviews human access quarterly has a second, faster-decaying population that the same cadence must reach.

The documented failure the review must prevent is the PlainID anecdote: an engineer grants an assistant broad read access, changes teams four months later, and the quarterly access review certifies the app registration twice, because the review certifies the account rather than the purpose-to-scope match. Nobody in the affected department knew the system existed. The lesson is specific: a review process that samples human accounts or app registrations without checking each agent against its registered purpose will pass exactly the agents it should question.

What the review covers follows from the identity bundle. Skyfire's buyer guidance lists verifiable identity, accountable owner, bounded authority, and revocable access as the four attributes, and the quarterly review verifies each: the identity still authenticates as itself, the owner is current, the authority still matches the agent's purpose, and revocation paths remain tested. Omada's guidance adds the operational items: delegation visibility across sub-agents and systems, and decision evidence proving each action stayed within authorized boundaries.

Drift is what the review finds when runtime monitoring is weaker than intended. Omada documents the pattern: agents chain decisions, delegate across systems, and remain inside existing permissions while acting beyond what anyone intended. Nudge Security's guidance adds the inventory dimension: most enterprise environments contain agents operating without IT approval, discovered through OAuth grants, API keys, MCP server connections, and non-human identities, so each quarterly review reconciles the discovered population against the registered one.

The volume problem makes the review design harder than the human equivalent. 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. No review reads behavior at that rate; the review instead verifies that the runtime checks that see every action are functioning, which is the correct division of labor between continuous monitoring and periodic governance.

The regulatory frame supplies the cadence and the record. SR 11-7 expects model risk governance commensurate with impact, and a periodic review of the identity controls governing agents on investment data belongs in that governance record. SR 26-2 third-party guidance applies where agent platforms and identity infrastructure run with vendors, including whether the firm can independently verify scopes. The EU AI Act, for firms operating in the European Union, expects documented, risk-proportionate governance, and quarterly review outcomes are the document.

A well-run review produces four artifacts every quarter: the reconciled agent inventory with owners, the list of scope changes and their rationale, the drift findings closed during the period, and the exceptions granted with expiry dates. Those artifacts are also what a model risk committee, a third-party examiner under SR 26-2, or a board asks for, which is the practical sign that the cadence is serving governance rather than filling a calendar.

Frequently asked questions

What does a quarterly review of agent identity cover?

It covers each agent's scope against its registered purpose, the currency of its named owner, the visibility of its delegation chains, the functioning of per-action runtime checks, and the testability of revocation paths, following the four identity attributes Skyfire's guidance lists: verifiable identity, accountable owner, bounded authority, and revocable access. The review also reconciles the discovered agent population from OAuth grants, API keys, MCP connections, and non-human identities, per Nudge Security, against the registered inventory so nothing operates unseen.

Why can quarterly reviews certify the wrong thing if designed carelessly?

Because a review that certifies accounts rather than purpose-to-scope matches passes exactly the agents it should question. The PlainID anecdote shows the mechanism: an over-scoped assistant kept being certified by quarterly access reviews twice over, even after the engineer who granted it changed teams and nobody in the affected department knew it existed. Omada's drift pattern, agents chaining decisions and delegating across systems while staying inside existing permissions, is invisible to an entitlement-only review for the same reason.

How should a review handle the action volume agents generate?

It should not try to read actions; it should verify the controls that do. Lumos counted more than 450,000 agent actions in a single week at a company of fewer than 200 people, more than 2,000 per person, so no periodic process can observe behavior at that rate. The correct division of labor is continuous runtime monitoring with periodic governance, and the review confirms the monitors function, findings close, and evidence is retained, which is also what SR 11-7 and SR 26-2 governance records expect.

What artifacts should each quarterly review produce?

Four: the reconciled agent inventory with current owners, the scope changes made during the period with recorded rationale, the drift findings closed, and the exceptions granted with expiry dates and reviewers. Those artifacts double as the documentation the EU AI Act expects for firms operating in the European Union, and as the evidence a model risk committee or third-party examiner requests. A review that produces no retrievable artifacts has happened as a meeting but not as governance.

Sources

  1. - federalreserve.gov
  2. - federalreserve.gov
  3. - artificialintelligenceact.eu