Resolve Who Is Actually Acting
Before you can govern an action, you need to know who is behind it. Identity resolution folds an evidence bundle into one canonical actor, with a confidence score and the anomalies that explain it.
Illustrative product view
Why identity resolution comes first
An action arrives with scattered evidence - an email, a company, a token, an agent handle. Identity resolution (resolve_from_evidence) folds that bundle into one canonical ResolvedActor with an identity_confidence between 0.02 and 0.95, a band (high ≥ 0.75, medium ≥ 0.40, low < 0.40), linked identities, named anomalies, and an explanation. The whole engine is pure and deterministic, so the same evidence always resolves the same way.
The signals and anomalies
| Signal / anomaly | Effect | Meaning |
|---|---|---|
| email_mismatch | −0.30 | Email does not match the claimed identity |
| company_mismatch / cross-tenant | −0.30 | Company or tenant does not line up |
| personal_account on corp domain | −0.25 | Personal account where a corp one is expected |
| agent_without_owner | −0.30 | Agent has no accountable owner |
| company_match / email_domain_match | +0.25 | Positive corroboration |
From confidence to a band
- high (≥ 0.75) - the actor is well corroborated
- medium (≥ 0.40) - plausible but with some gaps
- low (< 0.40) - weak or contradicted identity, treat with caution
- Every band comes with the linked identities, anomalies, and an explanation
How it connects
Identity resolution feeds the decision: a low-confidence or anomalous actor lowers the Action Trust Score and can trigger step-up. It also underpins non-human identity governance via the agent roster and owner linkage, and the identity graph (resolve, graph, build_identity_context) links related identities across a company.
Frequently asked questions
What is identity resolution?
Identity resolution folds a bundle of scattered evidence - emails, company, tokens, agent handles - into one canonical actor with a confidence score, a band, linked identities, and named anomalies. AegisAI’s engine is pure and deterministic, so the same evidence always resolves to the same actor and confidence.
How is identity confidence scored?
Confidence runs from 0.02 to 0.95 and is moved by named signals: mismatched email or company, a personal account on a corporate domain, or an agent without an owner lower it, while a company or email-domain match raises it. The result maps to a band - high, medium, or low - with an explanation of the drivers.
Does it handle AI agents?
Yes. Agents are first-class actors in resolution, with owner linkage and a roster. An agent without a linked owner is flagged as the anomaly agent_without_owner, and an unrecognized agent as unknown_agent, so ownerless or unknown automation lowers confidence rather than passing unnoticed.
How does identity resolution affect a decision?
A low-confidence or anomalous actor lowers the Action Trust Score and can trigger step-up verification, because the platform is less sure who is really behind the action. Resolving identity first means every downstream decision is made with a clear, scored picture of the actor.
Stay in the loop
One practical finance briefing a week - new guides, checklists, and benchmarks.
Know who is behind every action
Resolve scattered evidence into one scored, explainable actor.
Talk to us