Skip to content
The Future of Agentic Cyber Defense is Here: Vega II is Live | Watch on Demand
Vegasoft Vord II
Blog

Entity Profiling: Every Agent Knows Who It's Looking At

TL;DRSecurity teams have tracked users and devices for years, but each tool saw only the data it ingested. Vega builds one profile per entity across every connected source, in place, for its agent swarms to use.

Blog window reading Entity Profiling: Agents Know Who's Behind Every Alert

TL;DR

  • Security teams have tracked users and devices for years, but each tool saw only the data it ingested. Vega builds one profile per entity across every connected source, in place, for its agent swarms to use.
  • Each profile covers an account, device, or domain and carries behavioral baselines, so detections flag what's abnormal for that user, host, or service, not anything unfamiliar.
  • Triage and investigation weigh each alert by the entity's role, ownership, and normal behavior, with a source behind every fact. The result is fewer false positives and less time spent working out who or what an alert is about.

An attacker who steals one session doesn't stay in one tool. The same identity shows up in your identity provider, your mail logs, and your EDR, each under a different name. Each alert hands you only a string. Before anyone can judge it, someone has to work out who is behind it and whether that entity matters here. Traditionally, an analyst pulls the directory record, checks the EDR console, matches the accounts to one person, and decides which source to believe when two disagree. The answer stays in their head, so the next alert on the same person starts from zero. Entity Profiling, now in early access on the Vega Agentic Cyber Defense platform, works out that answer once, keeps it current, and hands it to every agent and engineer.

Who an alert involves decides what it means

The same alert means different things on different subjects. A suspicious sign-in on a dormant contractor account and the same sign-in on a domain admin call for different severity and a different response. A credential-access hit on a build server is not the same event as one on a finance workstation. Practitioners know this without thinking. An agent has to be told.

Today it usually isn't. Each source keeps a good record of the people and machines it sees, under its own identifier, and none of them knows the entity standing behind all of them. Enrichment returns a richer string, static lookup tables freeze a snapshot that starts aging the day it lands, and detections match strings, so one person with two accounts reads as two subjects.

Significance is a property of the entity: its role, its privileges, what kind of asset it is, who owns it, and what it normally does. An agent that can read those facts can rate an alert the way your best engineer would. One that can't is scoring strings.

Engineers stop rebuilding who someone is for each new alert and start checking the agent's call against a record they can read themselves. Vega does the assembly, and the engineer brings the judgment, so the time on each alert goes to the decision instead of the lookup.

Three weekend alerts are really one person

It's Monday morning, and three alerts are waiting from the weekend: impossible travel on alice@mycorp.com, a new inbox-forwarding rule on alice@mycorp.onmicrosoft.com, and credential access flagged on LAPTOP-ALICE. Without a resolved entity layer, they read as three subjects and three separate triages, connected only if someone remembers the first alert while reading the third.

Entity Profiling has already merged the records behind them. Both addresses resolve to one account, and the laptop resolves to the device she owns, so all three alerts land on the same person. Resolution is rule-based and conservative: a display name never creates or merges an entity, and an identifier shared by too many records doesn't link them, because a false merge is worse than two separate entities.

Three alerts resolved to one account
Three alerts resolved to one account
Diagram resolving two email addresses and a laptop to one account
One account behind two addresses and a laptop, resolved before anyone opened the incident.

Two minor alerts form a pattern

When Vega's agents triage and investigate, they query the same entity tables your engineers use. For this account, the record marks her as a human user rather than a service account and lists her department, job title, manager, the AWS roles she has used, and the devices she owns. Entra ID reports her department as Enterprise Security, while an older Windows record still says Legacy IT. Entity Profiling keeps both values with their sources and shows the most recent one.

That changes how the alerts read together. A forwarding rule on a security team member's mailbox, followed by credential access on her own laptop, is one pattern on a high-value identity. The agent rates it that way because the record says who she is and what she owns, and it writes that reasoning into the verdict.

The entity record the agent queries
The entity record the agent queries
Entity record listing department, title, manager, AWS roles, and owned devices with their sources
The facts the agent weighs, from one query on one table.

A new country stands out from routine travel

Who she is covers half the question. The other half is whether this activity is normal for her. Entity Profiling builds behavioral baselines for each user, host, and service. They cover the countries each user signs in from and how many sign-ins a day is normal for them. They also cover which hosts each service account logs on to, and which cloud actions each AWS or Azure principal usually performs.

So the impossible-travel alert is judged against the countries this user actually signs in from, not a global rule. Travel that's routine for her doesn't read as an attack, and a first-time country does. Detections in the Vega Library read the same baselines to flag a first sign-in from a new country or a service account logging on to a host it has never touched.

The verdict cites the baseline
The verdict cites the baseline
Agent verdict citing the user's sign-in country baseline and entity attributes
The verdict cites the baseline and the entity attributes it relied on.

Every fact the agent used shows its source

A verdict you can't trace is a verdict you can't trust. Every attribute in an entity table carries the data sources that contributed it and the timestamp of the event that last set it. A rebuild of the table doesn't reset that time. When the agent rates this account as high-value, you can query the record and see that the department came from Entra ID and when Entra ID last reported it.

Values that no source has reported within their expiry window drop out of view until one reports them again, so a stale attribute never passes as current. Vega refreshes the tables on a schedule and flags any table that falls behind. On a signal or incident, the entity card shows the same record, with the sources it was resolved from and when the entity was last seen.

Entity card on an incident
Entity card on an incident
Entity card on an incident showing identifiers, details, and the sources each value was resolved from
Who this is, where each value came from, and how current it is, without leaving the incident.

Built from every source, in place

A profile is only as complete as the data behind it. One built inside a SIEM covers only what was ingested there, so a sign-in in the IdP, a laptop in EDR and a role in the directory seldom meet in one record. Vega reads each connected identity, endpoint, cloud and DevOps source where its data already lives, whether that's object storage, a data lake or the SIEM you keep. Raw logs are never moved or re-ingested. Vega keeps only the resolved records and refreshes them on a schedule, so your team has no pipeline to build.

One model of your environment for everyone

An agent that can't tell a contractor from a domain admin will misread the alert, however good its reasoning is. Entity Profiling gives Vega's agents and your engineers one model of your environment: who's who, what's normal for them, and where each fact came from. It adds entity context to the knowledge every Vega agent works from. It's in early access now. If you're on Vega, ask Vega Support to turn it on, then select the account on your last high-severity incident to see what the agent knew. If you're not, bring that incident's identifiers to a demo, and we'll show you the resolved record on your own data.

Book a demo · Explore Entity Profiling

Frequently asked questions

What is Entity Profiling?

Entity Profiling is Vega's entity model with behavioral baselines. It resolves every account, device, and domain into one record merged from your connected sources, so engineers, detections, and agents read the same context.

Does Entity Profiling copy or re-ingest my data?

No. Entity Profiling works on your data in place, where you choose to keep it. It queries the sources you've already connected on a schedule, every 4 hours for accounts and every hour for devices and domains. Vega builds and maintains the entity records and baselines, so there's no new pipeline for your team to build or run.

What happens when an identifier matches more than one entity?

Vega doesn't pick one for you. On a signal or incident, the entity card opens with a notice that several entities are behind the value and shows each match on its own card, so you review every one. A value with no match shows that there's no entity data for it, and you can take it to a query instead.

  • #entity profiling
  • #behavioral baselines
  • #alert triage
  • #AI investigation