SentinelOne Singularity Operations Center - Alerts and the One Feature (Automatic actions)That Earns Its Place

Before I start, two things sit at the top right of the console that I am not going to cover here, because I already have.
Purple AI. Nothing has changed since I wrote about it, so I am not rewriting it.
My short view: it is a query writer and a triage reader, it is right often enough to save you real time, and you have to know where it is wrong. If it is enabled for you it is worth having. I will say the UI is not the best way to use it — I get more out of it through my own MCP setup, and maybe we discuss that at the end of the series. For the basics:
AI security, which is Prompt Security. I have a whole series on that one:
Now let us get to what this article is actually about.
Alerts
In the older console there were two tabs, Alerts and Threats. If you want that background it is here:
In the new console there are multiple tabs, and SentinelOne calls the whole thing Unified Alert Management.

If you read their documentation they keep going on and on about why it is the best. I will give you my perspective instead and show you. If you ask me, I do not like this and I am not going to pretend otherwise. That is my preference, yours might be different, and that is fine. Let us start.
The tabs
All is the first one. Everything lands here — STAR rule alerts, threat alerts, everything SentinelOne generates. The base is the same as the old console, so read the article above for that part. The view has changed a lot though.

The rest, straight out of the product:
Endpoint — EDR and CWS alerts

Custom Alerts — alerts from your custom rules or from Library detections

Identity — Singularity Identity Security alerts
Cloud — alerts from CWS and CDS

Partner alerts — alerts from third-party integrations
A few notes on those.
Endpoints is everything from SentinelOne's own detection engines.
Custom alerts is your STAR rules, so every rule you created shows up here.
Identity needs ISPM enabled or you will see nothing — I covered ISPM here and this touches it only lightly:
Cloud is where your ephemeral servers and Kubernetes workloads land — Cloud Workload Security and Cloud Data Security. Those also appear in the All tab, so do not get confused when you see them twice.
What actually changed in the view
Alert name. Earlier the console showed you the detected file or the command line. Now it leads with an alert name, which is a bigger change than it sounds — it changes what you scan for when you open the queue.

You can assign an alert to yourself. There was no option for that before.

The verdicts got split up. This is the one worth knowing properly. There used to be true positive and false positive. Now both are divided into sub-verdicts, and here is the full list rather than a couple of examples:
False positive | True positive |
Benign | Advanced persistent threat |
Benign but suspicious | Benign |
System error | Benign but suspicious |
Undefined | Data exfiltration |
User error | Denial of service |
Exploitation tools | |
Insider threat | |
Malware | |
PUA Adware | |
Phishing attack | |
Policy violation | |
Ransomware | |
Unauthorized access | |
Undefined |
Does it make sense to me? Not really.
If you are building an XDR and you want the data to mean something downstream, maybe it does.
What I will say is this: pick a convention and stick to it across your team, because with nineteen options everyone drifts, and then your reporting is worthless. In my own work the one I reach for most is False positive / Benign but suspicious, because that is the honest label for dual-use tooling that is not malicious but is not nothing either.
Filters are at the top — alert name, asset name, severity and plenty more. Anyone who has used an EDR or a SIEM knows what to do there, so I am not going in deep.

Group by. This works on every tab and it is genuinely useful. Group by hash, alert name, file name and more.

Group by alert name and all the same alerts collapse together, so you can see immediately how many of the same thing you are dealing with. Pick a site, account or group first, then group.

The views
Top right there is a summary dashboard. Click it and you get graphs for whatever tab you are on — detections for that view.

Next to it you get list view and triage view.
List view is line by line, you click a line and the alert details open at the side.


Triage view gives you a small pane for the alert list and a big pane for the detail. The screenshots will make it obvious.

Click into an alert and the side panel gives you Overview, Timeline, Indicators, Notes, Graph and Raw data. If you combine that with the old series you already know most of it.

The new part is Raw JSON and OCSF evidence, and it is worth explaining rather than shrugging at.

OCSF is the Open Cybersecurity Schema Framework, an open standard for describing security events so that data from different vendors has the same field names and the same shape.
The whole Unified Alert Management page is built on it, which is exactly how a Check Point firewall alert and a SentinelOne endpoint alert can sit in the same list and be filtered the same way — the partner integrations map their events into OCSF classes on the way in.
So why do you care?
Two reasons. If you forward alerts to a SIEM or build automation on them, OCSF is the contract you are coding against, and it is stable across sources. And when a field in the pretty view looks wrong or missing, the raw JSON and OCSF evidence is where you find out what was actually ingested. That is its real use — not reading for pleasure.
Four things below the views

Policy takes you to the policy tab, detect or protect. Policy is broadly the same as before so read this part of the old series, and we will come back to what changed later:
Blocklist is what the name says. One good change: you could only blocklist by SHA1 before, and now SHA-256 is supported too.
Worth knowing the small print on that, because it bit people. SHA-256 blocklisting arrived in platform version S-24.4.5 and is on by default from S-25.1.1. But enforcement needs the agent to be new enough — Windows and Linux agents 24.2 or later, macOS agents 24.3 or later. So you can add a SHA-256 hash and have it quietly do nothing on an older agent. Check your agent versions before you rely on it.
Action logs is like the activity log, except scoped to actions taken on alerts and nothing else.

Automatic actions — the one I actually like, and the reason this article exists.
Automatic alert actions
This is the one feature here I would call a real change. You create rules that act on alerts matching conditions you define, instead of an analyst opening each one. When an alert comes in, SentinelOne checks your rules and runs the actions from the first match.

Which means the repetitive, predictable alerts handle themselves and your people spend their time on the ones that need a human.
What you need before you start
Full Unified Alert Management permissions across all alert types at the scope you are working in. View-only can look at rules but cannot change them — if the property dropdown looks empty when you try to build a condition, check this first.
You can only create rules at or below your own permission scope.
Nothing to do with agent version.
Ops Center only. Not available on self-hosted or on-premises.
The one restriction that will decide whether this is useful to you
It does not run on alerts from a site that has an MDR add-on — Singularity MDR, Wayfinder or Vigilance. That is deliberate, so your rules do not fight the actions their analysts take.
Go and count your sites before you plan anything around this. In my own estate roughly half the sites carry an MDR add-on, and those are the busier ones, so the feature reaches a smaller share of the noise than you would hope. And because it is per site, it tracks your licensing — drop Vigilance somewhere and that site becomes eligible, add it and your rules silently stop firing there with no warning.
Scopes, and the thing that trips everyone
Three scopes: Site, Account, Global.
Here is the part to remember. This feature does not inherit downward like everything else in the platform. Site rules beat account rules, and account rules beat global rules. SentinelOne treats it as an operational tool for the analysts actually triaging, not a policy pushed from above, so whoever is closest to the alert wins.
When you create a rule, two dropdowns set the reach: leave both blank for global, pick an account for account level, pick an account and a site for site level.
Ranked or Independent
Every rule is one or the other.

Ranked — competes for priority inside its scope. Only one Ranked rule runs per alert, the highest-ranked one that matches. Lower-ranked rules never even evaluate that alert. Can use any action.
Independent — always runs when it matches, whatever the Ranked rules did. General actions only.
Evaluation order, for each alert:
Ranked rules first, by scope — site, then account, then global — and within each scope by rank.
First match executes and that phase stops. Then Independent rules, same scope order, newest to oldest, and every match runs. If two Independent rules set the same field, the first one to set it wins, the later rule skips that action, and the later rule's other actions still run.
Now the trap, and it is not obvious from the documentation. The actions are split into groups:
General actions — Analyst verdict, Change status, Add note, Assign to. Both modes.
Mitigation actions — Kill, Quarantine, Un-quarantine, Remediate, Rollback, Add to blocklist, Add to exclusions, and Mark as Resolved. Ranked only.
Asset operations — Run script, Disconnect from network, Reconnect to network. Ranked only.
Other — Partner actions, Retrieve file. Ranked only.

Look at where Mark as Resolved sits. It is a mitigation action, so it is Ranked only. If you build a simple auto-close rule in Independent mode and reach for Mark as Resolved, it is not there. Use Change status instead, which is a general action and does the same job for this purpose.
My rule of thumb:
if all you are doing is setting a verdict, changing status and adding a note, use Independent. It cannot be pre-empted by some other Ranked rule you forgot about, and that matters — a Ranked rule that works today silently stops working the day someone adds a higher-ranked rule above it.
Building the conditions
You pick a property, an operator and a value. Add Row puts another condition in the same group. Add Condition Group starts a new top-level group.
Two things the documentation does not tell you, which you learn in about five minutes of fighting the UI:
Rows inside a group are ANDed. Separate top-level groups are ORed. The read-only Query box at the side shows you the combined query as you build, and that is where you confirm it. Mine rendered exactly like this:
( File Path contains ".jsonl" AND File Path contains "/.claude/projects/" )
OR
( File Path contains ".jsonl" AND File Path contains "/.codex/sessions/" )
There is no file extension property. I went looking for one. Use File Path contains .jsonl instead and it does the same thing. And enter your values without quote marks — the Query box adds those for display. If you type the quotes into the value, the preview comes back empty and you will spend ten minutes wondering why.
Use the Results preview before you save. It simulates your conditions against historical alerts over a time range you pick. It is a simulation, not a promise, but it is the difference between a rule you tested and a rule you hoped about. If the preview shows zero, something is wrong with your condition. If it shows thousands, your rule is too broad.

A real example: auto-closing the AI assistant session files
Here is one I actually built, start to finish, so you can copy the shape.
The problem. On-Write DFI keeps flagging the .jsonl session transcripts that AI coding assistants write — Claude Code under ~/.claude/projects/, Codex under ~/.codex/sessions/. It classifies them as Infostealer or Malware. They are conversation logs. They are plain data, they cannot execute, and across my estate they turned up on thirteen sites and twenty-four endpoints, on twenty-three separate days in a month, with not one true positive among them.
The rule:
Field | Value |
Name | AI assistant session transcripts (.jsonl) - auto-resolve |
Scope | Account rules, site left blank |
Execution mode | Independent |
Status | Disabled first |
Expiration | Never expire |
Conditions, as the Query box shows them:
( File Path contains ".jsonl" AND File Path contains "/.claude/projects/" )
OR
( File Path contains ".jsonl" AND File Path contains "/.codex/sessions/" )Actions: Analyst verdict = False positive / Benign but suspicious, Change status = Resolved, Add note explaining why.
And now the part that matters more than the rule. Look at what I did not match. Not /.claude/ on its own, and not /Applications/Claude.app. Only the .jsonl files.
That is deliberate. A .jsonl transcript is inert. An executable sitting in the same directory is not, and a tampered application binary looks exactly like a legitimate one in a path filter. So the app binaries, the .js files under working directories, and anything in world-writable temp paths all stay in the queue for a human. The rule takes the files that cannot hurt me and leaves everything that can.
If you take one thing from this section, take that. Write the narrowest condition that still catches the noise, and make sure nothing executable can match it. It is very easy to write a rule that quietly closes the one alert you needed to see.
[Screenshot to add: The finished rule in the Account rules table (auto-actions-rule.png)]

Managing them
Get to the rules from the Alerts page, Automatic actions — or Policy and settings, General, Automatic actions. They are listed in three tables by scope, and a banner reminds you of the evaluation order.

The columns worth watching:
Alerts actioned, which tells you whether the rule is doing anything at all;
Last match timestamp, which tells you whether it still fires;
Version, which increments on every edit;
Query, so you can read the condition without opening the rule.
Check Alerts actioned a week after you enable something. If it is zero, your condition is too tight. If it is enormous, go and look at what it closed.
One honest limitation
There is no API for this. The rules are console-only — no REST endpoint, nothing in the alerts GraphQL schema beyond note and trigger operations. So you cannot script rule creation or keep them in version control, and if you manage many tenants you are clicking through the same rule in each one.
Two more things
Event Search is Deep Visibility. It used to open in a new tab, now everything is merged, so clicking it opens Event Search in place.
Just to add, in the newer console, PowerQuery and Deep Visibility queries need to be entered in the same place. Previously, these were separate, as PowerQuery searches and Deep Visibility queries had different place to search.

There is an Open in full tab button that gives you a better view in a new tab, and that is all there is to say about it.

That is the lot. The truth is they made the UI worse, in my opinion.
But that is what they did to make the tool better overall, so I will leave it there.
I will meet you in the next article. Till then, keep learning.
- written from hands-on work with the console, By Dean
------------------------------------------Cyberengage---------------------------------------------------




Comments