Skip to main content

Autonomous hunting

With autonomous hunting on, Scout doesn't wait to be asked. It keeps a ranked hunt list of things worth looking for, starts the top ones itself within a daily budget, collects evidence on the hosts it hits, contains them where a response policy allows it, checks the fix worked, hunts again to confirm, and ends every hunt with a result. Everything it does is listed in Scout activity, and an isolation it ran can be undone in one click.

On this page you see what Scout does on its own and what always waits for a person, how the hunt list is filled and ranked, how budgets work, and the difference between Shadow and On.

What Scout does on its own​

A typical autonomous hunt runs like this:

  1. Something worth hunting arrives. For example, a threat report names an indicator that touches one of your hosts. Scout adds it to the hunt list.
  2. Scout starts a hunt. Next time it checks the list, if the idea scores high enough and there is budget left, Scout opens a hunt, plans it and runs the queries.
  3. Scout collects evidence on the hosts it hits. Process trees, hashes and network connections from the endpoint, before anything changes.
  4. Scout proposes containment. If an active response policy covers the endpoint and the evidence is strong enough, the isolate, kill or quarantine runs straight away under that policy. If not, it waits for a person, as it does today.
  5. Scout proposes account and network actions, such as suspending a user or revoking their sessions. These are always one-click proposals for a person. See Connector actions.
  6. Scout checks the fix. Each action that supports it runs a check on the endpoint afterwards, for example that the endpoint really is isolated.
  7. Scout hunts again. It re-runs the queries that hit, scoped to the affected hosts and accounts, and records whether the activity is clean or still present.
  8. Scout records a result — Found, Nothing found or Inconclusive — and, when it found something, proposes a detection that starts in shadow.
  9. The morning summary lists what Scout hunted, found and did, with undo links, and what needs you.

What always waits for a person​

Autonomous doesn't mean unlimited. These never happen without someone clicking:

Never automaticWhy
Any change to an endpoint that no active response policy coversWithout a policy, Scout's proposals wait for approval exactly as in a supervised hunt.
Endpoints that don't carry one of the policy's tagsA policy only ever acts on endpoints with its tags. Everything else falls back to approval.
Account and network actions (suspend a user, revoke sessions, deactivate an access key, contain a host in another product, block an IP)They are one-click proposals. No response policy can cover them.
Patching, and any script Scout draftsPatches are proposals an organisation admin approves. A script Scout drafts stays unpublished until an organisation admin reviews and publishes it. See Patching.
Deleting a file, arbitrary commandsNot available to Scout.
Opening a live session on an endpointOnly you, with Connect, or Scout under its own autonomy rules can start one, and never on page load.
Turning a detection onDetections Scout proposes start in Shadow. A person decides when one goes On.
Your verdict on a huntScout records a result; the verdict of record is still yours. See Conclude a hunt.

The hunt list​

The hunt list is a ranked queue of hunt ideas for your organisation. Scout adds to it from several sources:

SourceWhat adds an idea
New threat intel that touches your assetsNew indicators or threat reports that match your assets. The more of your assets they touch, the higher the idea ranks.
Techniques you have no detection forATT&CK techniques that no live detection covers. See Coverage.
Patterns in the entity graphPatterns in your entity graph that are worth a hunt, for example an entity that several separate signals point at.
Scheduled hunts from the libraryScheduled hunts that are due, such as a weekly look for new persistence.
Alerts triaged as suspiciousAlerts that triage judged suspicious or worse.
A requestIdeas you ask Scout to add in a chat. These are always allowed.

A hunt Scout starts says where it came from on its Brief, for example Started by Scout from new intel: followed by the idea's title.

Each idea has a title, a hypothesis, the entities it starts from (hosts, accounts, indicators), any ATT&CK techniques, and a score between 0 and 1. The score weighs how fresh the signal is, how much of your estate is exposed, how big the coverage gap is and how strong the signal is. Each row on the hunt list shows the parts.

Scout keeps the list free of repeats. An idea isn't added again while the same idea is queued or running, while an open hunt already covers one of its entities, or when the same idea was hunted in the last 72 hours.

The hunt list is in Inbox › Scout activity, under Hunt list: What Scout plans to hunt next, highest score first. See Scout activity.

Hunt list statuses​

StatusMeaning
QueuedWaiting for Scout's next check. Ideas stay queued for 7 days, then expire.
LaunchedScout started a hunt from it. The row links to the hunt.
ShadowAutonomous hunting is in Shadow mode, so the idea was recorded but no hunt started.
SkippedNot hunted, with the reason: Daily budget used up, An open hunt already covers it, Hunted in the last 72 hours, Score below the threshold, Autonomous hunting is off or Doesn't touch your assets.
DuplicateThe same idea is already on the list.
ExpiredQueued for too long without being hunted.

Hunt an idea now, or add one​

  • Hunt now on an idea starts its hunt straight away. A person starting a hunt this way isn't held to the daily budget.
  • Skip an idea you don't want hunted.
  • To add an idea, ask Scout in a chat, for example add a hunt for new scheduled tasks on the finance laptops to the hunt list.

Budgets​

Scout checks the list at a set interval. Each time, it takes the top queued ideas that meet the lowest score, up to the per-check limit, until the day's budget is used.

SettingWhat it limitsDefault
Hunts a dayHow many hunts Scout may start on its own each day, 0 to 50.5
Hunts per checkHow many it may start each time it checks the list, 1 to 10.2
How often Scout checks the listEvery 15 minutes up to every 4 hours.Every 30 minutes
Lowest score to start (0–1)Ideas scoring below it are skipped.0.5

Hunts also count against the daily lake scan budget your organisation already has for Huntbase Lake; there is no separate one.

When the day's hunt budget runs out, the ideas left stay Queued for the next day; nothing is dropped. Scout records once a day that the budget is used up, and the morning summary shows how much of each budget was used.

Shadow and On​

ModeWhat happens
Shadow — record what Scout would start, start nothingScout fills and ranks the hunt list and records which hunts it would have started, but it starts none. Use it to see what Scout would pick before you let it.
On — Scout starts huntsScout starts hunts from the list within the budgets.

Start in Shadow and watch Scout activity for a few days. When the ideas look right, switch to On.

Hunts Scout starts run without a person to stop for, so they behave like an Operator hunt: Scout runs the queries itself and drafts the outcome. A hunt doesn't contain anything on its own unless a response policy allows it. See Choose how much Scout does.

Turn autonomous hunting on​

  1. Open Settings › [Organization] › Capabilities and find the Autonomous hunting card.
  2. Turn on Scout may start hunts on its own. Off, Scout only hunts when someone asks.
  3. Under Mode, choose Shadow or On, and choose How often Scout checks the list.
  4. Under Limits, set Hunts a day, Hunts per check and Lowest score to start, then save them.
  5. Under Where hunt ideas come from, tick the sources Scout may use.

Only org admins can change these settings; other members see them read-only.

Scout starts hunts on behalf of the person who turned hunting on, or who switched it from Shadow to On, with that person's access. The card says who, for example Scout acts on behalf of Sam Patel. It changes to whoever next turns it on, or switches it from shadow to on. If that person has left the organisation (the card then reads a former member) and nobody else can be named, Scout skips the launch and says why in Scout activity.

To have Scout act on endpoints as well as hunt, set up a response policy. Without one, autonomous hunts still find, collect and propose, and every change waits for a person.

Stop Scout​

  • To stop new hunts, turn off Scout may start hunts on its own, or switch the mode to Shadow. Hunts already running carry on until they settle.
  • To stop Scout acting on endpoints at once, turn the response policy off. It takes one org admin and applies straight away.
  • To stop Scout changing one hunt for a while, pause Scout on that hunt.

Next steps​