In this section

Device Control, Network Protection, SmartScreen and Application Control in Defender for Endpoint

Module 0

Introduction

Attack surface reduction rules and antivirus act on what a process does and what a file is. The controls in this lesson act on what a device may connect to: a USB drive, a website, a binary that wants to run, a process that wants to behave in a way it normally never would.

Module 6 treats them as one surface, because an attacker who meets one of them routes around it through another, and a policy written for one says nothing about the rest.

This lesson reads the estate's records for five of these controls. Two have a clear history in the data, one has a single policy firing on a single laptop, and two have no records at all. Each of those is a different finding, and the work of this lesson is telling them apart.

1. Nothingthe activity happens and may be recorded as activity2. Auditthe control records what it would have refused, and allows it3. Blockthe control refuses, and records the refusal4. Reviewedsomeone reads the records and adjusts the policy

Every control in this lesson sits at one of these stages on each device.

The figure is the scale every control here gets placed on. A control with no policy leaves activity to happen; in audit it records what it would have refused and allows it; in block it refuses. The last stage, a person reading the records, is the one that makes the other three worth having.

01

Seventy-Seven Mounts

What visibility looks like without control

The sensor records a USB drive being mounted whether or not any policy applies to it. Across the estate in the period:

DeviceEvents
| where ActionType == 'UsbDriveMounted'
| extend Product = tostring(parse_json(AdditionalFields).ProductName)
| summarize Mounts = count(), Devices = dcount(DeviceName), Products = dcount(Product)

The count of mounts, devices and drive models comes back as one row:

Mounts  Devices  Products
------  -------  --------
77      15       4

77 mounts on 15 devices, from 4 different drive models. That is ordinary business: people move files on USB drives, and an estate that has never looked will find this much without trying. Against that, the device control policy fired 4 times, all on one laptop, all on the last day of the records.

So the estate can see removable storage being used and almost never controls it. A mount record is visibility: it says a drive arrived and which letter it took. It says nothing about what was read from it, what was written to it, or whether anything was refused. Device control is the part that decides, and on this estate it decided four times.

That difference matters most when a question arrives about data leaving. A mount record proves a drive was present at a time, and the investigation then has to reconstruct what happened from file events, if there are any. A device control record proves a specific operation was attempted and what the policy did about it.

One is a starting point; the other is an answer. An estate that wants answers to that question has to decide in advance which operations it will control, because device control records only what a policy was written to see.

Module 6 opens with device control as one switch over every device family, because USB storage is only the most familiar of them. Printers, portable devices and optical drives are all in scope, and a policy that covers only USB drives leaves the others as open as the mounts above.

It is worth being exact about what the mount records are good for even without a policy. They are the inventory a device control policy is written from: which devices use removable media at all, how often, and which drive models the business actually owns.

A policy that allows the company's approved drives and refuses the rest needs exactly that list, and on this estate it can be built from the records already here.

02

A Write Refused

One block-write policy on one laptop

The four device control records are on NE-DAVIS-LT, on 15 March. Lined up with the drive mount that came before them:

DeviceEvents
| where DeviceName == 'NE-DAVIS-LT'
| where ActionType in ('UsbDriveMounted', 'RemovableStoragePolicyTriggered')
| where Timestamp > datetime(2026-03-15)
| extend F = parse_json(AdditionalFields)
| project Time = format_datetime(Timestamp, 'HH:mm:ss'),
    Event = iff(ActionType == 'UsbDriveMounted', 'mount', 'policy'),
    Item = coalesce(FileName, tostring(F.DriveLetter)),
    Policy = tostring(F.RemovableStoragePolicy)
| order by Time asc

The mount and the four policy records, in time order:

Time      Event   Item                 Policy
--------  ------  -------------------  --------------------------------
10:11:42  mount   F:
10:12:30  policy  Q1-pipeline.xlsx     Block write to removable storage
10:12:30  policy  Q1-pipeline.xlsx     Block write to removable storage
10:14:02  policy  Accounts-North.xlsx  Block write to removable storage
10:14:05  policy  Accounts-North.xlsx  Block write to removable storage

A Seagate Expansion drive was mounted as F: at 10:11. Within three minutes, Explorer tried to write two spreadsheets to it, Q1-pipeline.xlsx and Accounts-North.xlsx, and each attempt fired the policy named Block write to removable storage. Each file appears twice, a few seconds apart at most, which is consistent with a copy that failed and was tried again.

Read the record for what it contains and what it does not. It names the policy that fired, the media and the bus type, and the file. The full device control record also carries the access requested and the policy's verdict, which these records do not include, so the policy's name is doing the work here.

Module 6's lesson on monitoring covers reading the verdict directly, and why a policy name is not a substitute for it. A policy can be named Block and configured to audit, and the name will say nothing about the mismatch.

Notice the process as well. The writes came from explorer.exe under the laptop's user, which is what a person dragging files onto a drive looks like. A tool copying files in bulk would show a different process and a different rhythm, and both would be refused by the same policy.

There is also a ceiling to know about. Microsoft's report documentation says advanced hunting keeps at most 300 of these events per device per day, and points to the device control report for the rest. A laptop with a noisy policy can run out of records in a busy afternoon, and a query will then show less than happened.

The finding for the estate is simple. One laptop has a policy that refuses writes to removable media; the 15 devices in the mount records show no sign of one. The policy exists, because it fired; what the records show is how far it reaches.

Whether that is a pilot or an accident, the records cannot say, but either way the estate's exposure to data walking out on a USB drive is set by the other devices.

03

Audit, Then Block

Network protection read from its records

Network protection is the control that refuses connections to addresses with a poor reputation, or to ones an administrator has listed. Every record it produced in the period:

DeviceEvents
| where ActionType startswith 'ExploitGuardNetworkProtection'
| extend F = parse_json(AdditionalFields)
| project Day = format_datetime(Timestamp, 'MM-dd HH:mm'), DeviceName,
    RemoteUrl, Browser = InitiatingProcessFileName, IsAudit = tostring(F.IsAudit)
| order by Day asc

The seven records, oldest first, with the mode each was produced in:

Day          DeviceName     RemoteUrl               Browser      IsAudit
-----------  -------------  ----------------------  -----------  -------
03-04 10:21  NE-COLLINS-LT  ne-sharepoint-docs.com  chrome.exe   true
03-04 10:24  NE-DAVIS-LT    ne-sharepoint-docs.com  chrome.exe   true
03-06 14:02  NE-FOSTER-LT   files-transfer-hub.net  firefox.exe  true
03-09 11:37  NE-ZHANG-LT    files-transfer-hub.net  firefox.exe  true
03-12 09:15  NE-LEWIS-LT    ne-sharepoint-docs.com  chrome.exe   false
03-13 16:44  NE-FOSTER-LT   files-transfer-hub.net  firefox.exe  false
03-14 08:52  NE-KHAN-LT     ne-sharepoint-docs.com  chrome.exe   false

The first 4 records have IsAudit true, and the last 3 have it false. The change falls on 11 March: before it, network protection on these laptops was in audit, and after it, in block.

The browsers are Chrome and Firefox, never Edge, for a reason the next section explains. Two destinations appear in both phases: ne-sharepoint-docs.com, a phishing site dressed as the estate's own document store, and files-transfer-hub.net, a file sharing site on the administrator's own block list.

Audit mode needs reading carefully. Microsoft's description is that audit logs whenever a user connects to an address that would otherwise be blocked.

The connection is recorded, not refused. So on 4 March two users in Chrome reached the phishing page, and on 6 and 9 March two users in Firefox reached the file sharing site. The records of those visits look like detections, and in a dashboard they are counted with the blocks, but nobody was stopped.

The blocks after 11 March are the control working. Three users went to the same two sites and were refused. That is the payoff of audit first: a week of records showed the policy would refuse nothing the business needed, and the move to block was made on evidence.

Had a legitimate site appeared among the audited destinations, the right response would have been an allow indicator before the change, not a delay to it; Microsoft's documentation notes that allow indicators take precedence over blocks. It is the same discipline lesson 0.4 applied to ASR rules, and it is cheap to follow.

The estate's records also show its cost: every day in audit is a day the control is recording visits it was written to stop, so audit is a phase with an end date, not a resting state.

The question audit leaves behind is what happened during the audit week. Two users reached a page built to collect credentials, and the records do not say what they did there. Module 6 does not investigate that, but the engineer who moved the control to block is the one who knows it happened, because only the audit records say so.

04

Two Browsers, Two Controls

Edge relies on SmartScreen, the others on network protection

The phishing domain appears in a second kind of record. Every event for ne-sharepoint-docs.com, grouped by the browser and the control that produced it:

DeviceEvents
| where ActionType startswith 'ExploitGuardNetworkProtection'
    or ActionType == 'SmartScreenUrlWarning'
| where RemoteUrl == 'ne-sharepoint-docs.com'
| summarize Events = count(), Devices = dcount(DeviceName)
    by Browser = InitiatingProcessFileName, ActionType

Grouped that way, the records split by browser:

Browser     ActionType                            Events  Devices
----------  ------------------------------------  ------  -------
msedge.exe  SmartScreenUrlWarning                 3       3
chrome.exe  ExploitGuardNetworkProtectionBlocked  2       2
chrome.exe  ExploitGuardNetworkProtectionAudited  2       2

The Chrome records are network protection. The Edge records are SmartScreenUrlWarning: 3 warnings, on three laptops, for the same site. Microsoft's network protection documentation is explicit that on Windows network protection does not monitor Microsoft Edge. Edge relies on SmartScreen, and network protection exists to extend the same kind of protection to Chrome, Firefox and processes that are not browsers at all.

That has two consequences for an engineer. First, reporting on web protection needs both record types, or 3 of the campaign's 7 records disappear from the report depending on which browser each user preferred. Second, the two controls are configured separately. Microsoft states that for Chrome, Firefox and non-browser processes, network protection must be in block mode to apply custom indicators and web content filtering.

For Edge, SmartScreen has to be enabled. An estate can have one right and the other wrong, and the records are how to tell: a browser whose visits to a known bad site produce no record from either control is a browser no control is watching.

A warning is also not the same as a block. SmartScreen in Edge shows the user a page saying the site is unsafe, and depending on configuration the user may be able to continue past it.

The warning records here say the page was shown; they do not say what the user did next. Module 6's web filtering lesson covers the setting that removes the choice, and the records that show whether users overrode a warning.

The practical rule is to report web protection by destination, not by control: every record for a domain, from every control, on every browser. That is the shape of the query above, and it is the only shape that shows a campaign whole.

Seen that way, the phishing site reached seven laptops through two browsers over ten days, was allowed on two of them, warned on three and refused on two. No single control's report would have said that, and it is the sentence a security lead actually needs.

05

The Records That Do Not Exist

Application control and exploit protection

Two controls in Module 6 leave no trace on this estate. The test is a count of each family of records, with ASR and network protection alongside to prove the query can find what exists:

DeviceEvents
| summarize
    AppControl = countif(ActionType startswith 'AppControl'),
    ExploitProtection = countif(ActionType startswith 'ExploitGuard'
        and not(ActionType startswith 'ExploitGuardNetwork')),
    NetworkProtection = countif(ActionType startswith 'ExploitGuardNetwork'),
    Asr = countif(ActionType startswith 'Asr')

The four counts come back side by side:

AppControl  ExploitProtection  NetworkProtection  Asr
----------  -----------------  -----------------  ---
0           0                  7                  134

Network protection has 7 records and ASR has 134, so the query is capable of finding control records. Application control and exploit protection have none.

Either those controls are not configured, or they are configured in a way that has never fired on these devices; on a month of activity that included a ransomware binary, the first is far more likely, and Module 6 shows how to confirm it from configuration rather than from silence.

What would application control have decided? notlocker.exe ran from a user's Temp folder, started by PowerShell. QuoteGen.exe, the add-in the ASR pilot broke in lesson 0.4, runs from Program Files. An application control policy allows what its rules list, by signer, by path or by how the file arrived, and an unknown binary in a user's Temp folder is listed nowhere.

The ransomware would have been refused before any antivirus verdict existed, at the moment PowerShell tried to start it, and QuoteGen would have needed an explicit rule. That trade, a list to maintain against a whole class of threat refused, is what Module 6's three application control lessons are about.

Exploit protection works at a different level: it changes how a specific process behaves in memory, so that a technique an exploit depends on fails. Nothing on this estate exercised it, and nothing records whether it is set. It is the hardest control here to judge from records, which is why Module 6 teaches it as a configuration to verify rather than a record to wait for.

The method this section used is the general one for any absence. Count the family you expect to be empty beside families you know exist, in the same query, over the same window. If the known families return their records and the expected one returns zero, the zero means something. If everything returns zero, the query is wrong, and nothing has been learned about the estate.

06

What Module 6 Builds On It

One control surface

The estate's position on these five controls fits on one card:

Where each control stands on this estate

Five controls, three stages

Device control

one block-write policy, seen firing on one laptop

Network protection

audit until 03-11, block after: Chrome and Firefox

SmartScreen

warnings in Edge for the same phishing site

Application control

no records at all

Exploit protection

no records at all

Module 6 starts with device control across every device family, then printers, then application control in three lessons: AppLocker as Microsoft itself describes it, the defaults that are not what people assume, and trust that depends on how a file arrived.

It covers exploit protection and the web content filter, the classification that has to happen before a policy can decide, and four innocent reasons a query returns nothing. It finishes by writing the policy and running one intrusion against ten controls.

The reason for treating these as one module is the route an attacker takes when one control holds. A user who cannot copy files to a USB drive may upload them to a file sharing site, which is network protection's problem. A binary that application control refuses may be replaced by a script, which is ASR's.

A phishing page Edge warns about may be opened in Chrome, which is network protection again. Lesson 0.10 runs one intrusion through all of them on this estate, and the card above predicts where it gets through. Each control closes one route, and the engineer's job is to see which routes are still open on which devices, which is what the card above does for this estate.

What Module 6 builds

Module 6

6.1 and 6.2

one switch for every device family, and a printer that writes to disk

6.3 to 6.5

AppLocker as it really is, application control defaults, trust by how a file arrived

6.6 and 6.7

exploit protection on a process, and a web filter that reads a name off the wire

6.8 to 6.11

classify before deciding, empty queries, writing the policy, one intrusion and ten controls

The habits to carry into it are the ones this lesson used. Separate visibility from control: a mount record is not a device control decision, and an inventory built from mounts is where a policy starts. Read the mode of every record, because an audited connection went ahead.

Report by destination across every control and browser. Prove a query can find something before reading its silence. And place each control on the stage scale for each device, because an estate is rarely at one stage everywhere.

Each of those habits came from a record on this estate, and Module 6 uses the same records to show each control's settings changing what they say. A setting that is saved and changes nothing in the records has not yet been shown to work.

Do

Practice

Decide which visits a control stopped
prepared evidence

The prepared evidence below is every network protection and SmartScreen record on the estate in the period. Decide which connections went ahead, which records came from the browser network protection does not cover, and what the audited custom block list entries say about indicators. Each answer comes with its reasoning, so check it against the rows rather than only the choice.

Read across each row before deciding: ActionType says which control produced it, the browser says why that control and not the other, and IsAudit says whether the user was stopped. Only a network protection record with IsAudit false is a refusal; a SmartScreen warning is a page shown to the user, and the record does not say what they chose next.

The dates matter as well. Sort the rows by time and the move from audit to block on 11 March is visible as a line through the table, with the same two destinations on both sides of it.

Everything above the line is the cost of audit; everything below it is the control doing its job. Count the users above the line: they are the ones whose visits deserve a follow-up question, because nothing stopped them.

Then run the device control query on this page without the date filter and without the device filter, and count how many devices mounted a drive on the same day as a policy record.

On this estate the answer shows how narrow the one policy's reach is. Then list the devices with mounts and no policy record at all; that list is the starting scope for the device control work in Module 6, and it comes from records the estate already keeps.

Finally, change the last query's AppControl filter to a family you expect to be empty, prove it returns zero, and then add one you know exists to prove the query still works.

Device control, network protection, SmartScreen, application control and exploit protection are one surface, and on this estate they sit at every stage from nothing to block. Visibility is not control, audit is not refusal, a warning is not a block, and an absent record is a finding only once the query has been proved able to return one.

The next lesson moves from preventive controls to detection: what the EDR sensor does beyond its defaults, and what an estate gains by changing them.