| Agent | SentinelOne on every line. The detection lines carry the Complete agent, which adds the deeper telemetry the hunting and correlation work depends on. |
|---|---|
| Autonomous action | Behavioral models run on the device itself. Kill, quarantine, and network isolation happen at machine speed and do not wait for the desk, a console session, or your connectivity. |
| Correlation | Detection telemetry is forwarded into Fluency, where it is joined to identity, mail, server, and network sources so an analyst sees the sequence rather than one alert. |
| Who acts, Fortify-MDR | The desk monitors around the clock, triages, and raises what matters to you with the recommended action. Containment on the endpoint is the agent's; anything further is yours to authorize. |
| Who acts, Fortify-XDR | Adds cross-layer hunting and automated orchestration across the connected sources. The desk drives the investigation and hands you a finished narrative. |
| Who acts, Fortify-XDR+ | Adds hands-on cleanup: our own analysts remediate the identified events rather than sending you a task list at two in the morning. |
| Cover | Round the clock, weekends and public holidays included. |
| How it bills | Monthly, against each endpoint. An endpoint here means a workstation or a server. |
| What is deployed | The SentinelOne Kubernetes agent, running as a workload inside the cluster, covering containerized and cloud-native services rather than the developer laptops that build them. |
|---|---|
| Tiering | The same three tiers apply, in the same order: monitored, extended, and remediated. |
| How it bills | Monthly, against each Kubernetes agent. Agent count follows the shape of the cluster, so we count it with you and confirm the number in writing before it reaches an invoice. |
| Where it stops | The cluster lines protect the cluster. Workstations and servers outside it need the endpoint lines above. |
Prevention has a ceiling, and it is lower than the brochure says
Every preventive control eventually meets an attacker who does not need to break it. Stolen credentials do not trigger malware detection, because nothing malicious runs. Administrative tools already present on the machine do not look like an intrusion, because they are the same tools your own engineers use. The category of attack that ends in a ransom note usually contains no malware at all by the time it matters.
The outcome is not decided by whether something got in. It turns on the gap between the first unusual action and a competent person looking at it. Squeeze that gap down to minutes and you have an incident. Leave it at weeks and you have a disclosure obligation, a forensics invoice, and a difficult call with your insurer.
What the agent is permitted to do without asking
The SentinelOne agent builds a running model of process behavior on the device and judges sequences rather than files. It has standing authority to act on that judgment: terminate the process tree, quarantine what produced it, and cut the machine off the network while leaving our management channel open so the investigation can continue.
Two properties matter more than the feature list. The judgment happens locally, so it still happens when the site is offline or the link to the cloud is saturated. And the agent keeps the chain of what led to what, so the analyst inherits an account of the incident instead of a timestamp and a file name.
Where the account of an incident is assembled
Taken alone, an endpoint alert is a fragment. It reports that a process was killed on one machine at 03:12 and says nothing of the sign-in from an unfamiliar address forty minutes before, or the mail rule created quietly afterwards.
Telemetry is forwarded into Fluency, where endpoint detections are correlated with identity, mail, server, and network sources. That is what turns six unrelated notifications into one sequence with a beginning. It is also what makes the difference between telling you an endpoint was cleaned and telling you which account was used, what it reached, and whether it is still holding a session.
The three tiers are a question about who pulls the trigger
The tiers differ far less in what they can see than in who is permitted to act on what is seen, which is the decision most buyers are actually making.
- Fortify-MDR watches continuously and brings you findings with a recommendation. Suits organizations with their own capable IT function that wants the eyes but keeps the hands.
- Fortify-XDR adds cross-layer hunting and orchestrated response across the connected sources, so the investigation runs to a conclusion rather than stopping at the endpoint boundary.
- Fortify-XDR+ goes further again: our own analysts perform the cleanup on identified events. Suits organizations whose honest answer to a 2am page is that nobody will read it until morning.
Nothing prevents mixing tiers. Domain controllers, finance workstations, and the machines that touch payment systems commonly sit at the higher tier while the rest of the estate sits below it.
The isolation happens first, and the argument happens afterwards
When the agent decides a machine is doing something it should not, it disconnects that machine and tells us. It does not open a ticket and wait for business hours. A false isolation costs one user part of a morning. A late isolation costs the estate.
The desk then reviews every containment, releases what was wrong, and carries the ones that were right through to a finished account of what happened. That review is the part you are actually buying: software alone produces a queue nobody drains, and a queue nobody drains is indistinguishable from having bought nothing.
Where this Article stops
- Detection is not prevention. Stopping unapproved software from running at all belongs to Article II, and it is the cheaper half of the problem.
- Coverage follows the agent. A machine nobody enrolled is not watched, and neither is a cloud service compromised without any endpoint being involved unless its logs are being forwarded into the correlation layer.
- Phones and tablets are not endpoints for the purposes of this Article. Mobile detection is the Zimperium line in Article IV.
- Response contains and cleans. It does not put deleted data back. That is Article VI, and no quantity of detection is a substitute for it.