Anomaly Detection

Aegis watches its own activity record for unusual behaviour — off-hours bursts, volume spikes, bulk exports, unfamiliar locations — and lists each flagged pattern for a person to judge.

Most of Aegis records settled facts: an approved policy, an accepted risk, a control that passed its last test. This screen is different. An anomaly (a flagged deviation worth a human look) is raised when activity stands out from the normal pattern — dozens of actions outside typical working hours, an hourly count far above the usual baseline, a run of exports from someone who rarely exports.

What this module actually watches

The signals come from Aegis's own audit trail — actions taken inside the platform — not from firewalls, sign-in providers or a threat feed. Aegis is not a SIEM (a security information and event management system) and does not replace your security tooling. Its job is narrower: frameworks such as ISO 27001 and NIS2 (a European cybersecurity law) expect evidence that unusual activity is reviewed, not only logged.

Who uses it

Anomaly Detection is a licensed feature, carried by the Professional and Enterprise licences. Where it is not enabled, a centred notice under an amber icon replaces the list: Anomaly detection is not available on your current plan, with an invitation to upgrade. Reading it is open to every role, Viewer and Contributor included — they can open any row and ask for the AI explanation. Changing a review status or using the Mark false positive shortcut is drawn only for Manager and Admin. The screenshot below was captured as a Contributor — hence the plain Review Status badges and no Mark false positive button.

What's on this screen

The header reads Anomaly Detection, with "Monitor and investigate detected anomalies across your compliance environment." underneath. There is no create button — anomalies are detected for you, never entered by hand — and no export. Below it sit four cards: Total Anomalies, Critical in red, Warning in amber and Unacknowledged in blue. The captured screen reads 12, 5, 4 and 6 — a seeded demonstration set, so treat the numbers as illustration. If they fail to load, a red notice replaces the cards with a Retry button; the table still works.

Under the cards are two dropdowns — severity (All Severities) and review status (All Review Statuses). They are the only way to narrow the list; there is no search box or date picker on this screen. (The search field in the top bar searches Aegis as a whole, not this table.)

The rest is the table: Severity (a coloured badge), Type, Description (a line such as "Action volume of 38 actions/hour exceeds baseline of 6.4 (z-score: 7.52)" on the captured screen, where a z-score measures how far activity sits from the baseline), User (System for platform activity, or Unknown user when the account can no longer be resolved), Detected and Review Status. Descriptions are shown in your interface language. Rows run twenty to a page, newest first; a Manager or Admin sees one more, unnamed column carrying the Mark false positive button.

  1. Select the ? button in the top bar. A help panel slides in with short topics for the page you are on and a search over the bundled help.
  2. Select the open-book button beside it. It opens this guide in the app at the chapter for the screen you are on.
  3. Check the name at the top right — the account you are signed in as decides the rest. A Manager or Admin gets the write controls; everyone else reads the same list without them.
  4. Open the Risks group in the left menu and select Anomalies. /anomalies loads with its cards, filters and table — it sits with the risk register because Aegis treats detection as a risk signal, not a security console.
The Anomaly Detection list seen as a Contributor: summary cards, the two filters, and the table with read-only status badges — /anomalies.
The Anomaly Detection list seen as a Contributor: summary cards, the two filters, and the table with read-only status badges — /anomalies.

Read and triage the list

  1. Read Critical for how many flags sit furthest from the baseline, and Unacknowledged for how much of the total nobody has judged.
  2. Scan the table from the top — newest first. Description states what was found, including the counts and baselines behind it.
  3. Check User and Detected to place the activity. A late-night burst from an account that normally works office hours deserves more attention than the same pattern from a known overnight job.
  4. Open the severity dropdown and choose Critical. The table reloads with only critical flags; work down through Warning and Suspicious after.
  5. Open the review-status dropdown and choose New for the untouched queue, or Investigating, Confirmed or False positive to see what is in progress or decided.

Open an anomaly to see what triggered it

  1. Select anywhere on a row. A dialog titled Anomaly Details opens, showing Type, Severity, Review Status and Detected in a four-field grid.
  2. Read Description below the grid — the full detection line, without the two-line truncation the table applies.
  3. Read Related Events: the audit-log entries that triggered this anomaly, each with its action and timestamp. Where an anomaly carries none — older rows detected before this link existed — the section says so rather than showing an empty list.
  4. Select View in audit log beside an event to open that entry in the Audit log.

Work an anomaly through its review status

Every anomaly carries a review status (where it sits in your triage): New, Investigating, Confirmed or False positive. This is the trail an auditor reads. For a Manager or Admin, the Review Status cell is a dropdown rather than a badge.

  1. Open the dropdown in the row's Review Status cell. It offers only the moves allowed from that row's current status, so an illegal step is never presented. From New you can go to any of the other three; nothing moves back to New.
  2. Choose Investigating when you start looking into a flag. It saves at once, a Review status updated confirmation appears, and the table and cards refresh.
  3. On a verdict, set Confirmed if the activity is genuinely unusual, or False positive if it was expected. Either marks the anomaly acknowledged, so the Unacknowledged card falls by one.
  4. To correct a wrong call, open the dropdown again: Confirmed can move back to Investigating or across to False positive, and the reverse. A refused move shows Failed to update review status and the row snaps back to the value Aegis holds.
A confirmed anomaly is not an incident

Marking an anomaly Confirmed records your verdict; it does not open a case, assign an owner or start containment, and there is no "promote to incident" button here. If a confirmed flag points to something serious, raise it yourself as a full incident and reference the anomaly there.

Mark an anomaly as a false positive in one click

Every row whose status is not already False positive offers a Mark false positive button at the right end — the fast path for clearing noise, for Managers and Admins only. It is a shortcut for choosing False positive in the row's dropdown, not a separate action: both save the same verdict and leave the same entry in the audit trail.

  1. Satisfy yourself the row is expected behaviour — read the description, user and time, opening the detail if you need the triggering events.
  2. Select Mark false positive. The table refreshes, the row's status reads False positive and the button drops off it. If the anomaly had not yet been acknowledged, the Unacknowledged card falls by one.
  3. To review what has been cleared, set the review-status filter to False positive; set it back to New for the live queue.
Cleared anomalies are kept, not deleted

A confirmed or false-positive anomaly stays in the record under its review status, so the history of what was flagged and how it was judged stays complete. There is no delete control: a wrong call is corrected by moving the review status, not by removing the row. Moving a decided anomaly back to Investigating also clears its acknowledgement, so it returns to the unacknowledged queue and the Unacknowledged card rises by one.

The AI assist

The detail dialog carries one AI action, Explain (AI), at the bottom right: a plain-language account of what happened, why it matters, and a proportionate response — for example, confirm with the user, review the related events, raise an incident, or mark it as a false positive. These are suggestions only. Any role that can open the list can run it.

  1. Open an anomaly and select Explain (AI). A dialog headed AI Anomaly Explanation opens, describing what it covers and noting that it is read-only.
  2. Select Explain anomaly. The text streams in as Aegis loads the record, assesses the deviation and severity, explains what happened and recommends a response; a confidence figure and the evidence used arrive at the end.
  3. Keep the briefing with Save as record if it is worth referring to later — saved explanations are listed under the anomaly in its detail dialog. Run again produces a fresh one.

Aegis reasons over that one anomaly's record — the recorded baseline and observed values, and the signals it computed itself — and nothing else. It reads no external data, invents no metrics or events, and changes nothing: the review status, the acknowledgement and any escalation stay a person's decision. Each run costs one AI credit.

What Aegis flags

Flags are produced in the background by comparing recent activity against a rolling baseline of normal behaviour. There is no rule builder or threshold editor in this release.

Type What it flags
Volume Spike An hour's action count far above the usual rate, with baseline and z-score.
Off-Hours Activity Actions taken well outside the typical working hours for the account, with the hours named in UTC.
Bulk Export An unusual number of exports, especially from someone who rarely exports.
Unusual Action An action type that is out of character for the account, named with how often it was performed.
Privilege Escalation Privilege-changing actions from an account that does not normally make them.
New Location / IP Actions from client IP addresses the account has not been seen using before.

Severity comes from the z-score alone — how far the activity sits from the baseline. Above four is Critical, above three Warning, above two Suspicious; anything closer to the baseline than that is not raised at all. A Volume Spike is only raised from Warning upwards. Those three are every severity the detector assigns, and all three are offered in the severity filter.

Tips and limits

Where this connects

The Audit log is the raw trail these flags are computed from; this page is the unusual-behaviour view of it. When a flag turns out to be a real security event, open Incidents. A recurring pattern belongs in the risk register, and a well-kept CMDB helps you judge an anomaly in context. Credit balances live on the AI dashboard; role limits in Part 3.