Machine Investigator

Type in a device name. Get a straight answer about whether that machine is in trouble — and what to do about it.

Preview feature   ยท   Requires Microsoft Defender for Endpoint with Advanced Hunting

The short version

Someone asks you “is there something wrong with Sarah’s laptop?” Answering that properly today means writing half a dozen Advanced Hunting queries, reading them in separate tabs, and holding the whole picture in your head.

Machine Investigator does that in one pass. You enter one device — or up to twenty-five — and each one comes back with a colour-coded verdict, the facts behind that verdict, and concrete remediation steps you can copy and paste.

Nothing is installed on the machine you are investigating. The data comes from the Defender sensor telemetry already in your tenant, so you can investigate a laptop that is switched off, in another country, or that you have never touched.

The verdict

Every device gets one of three answers, with a single sentence explaining why. This is the part you read first, and often the only part you need.

๐ŸŸข Healthy

Nothing significant found in the period you looked at. Defender is doing its job and there is nothing outstanding.

๐ŸŸก Needs attention

Real findings, but nothing on fire — unpatched high-severity CVEs, a handful of configuration gaps, or unusual activity worth a look.

๐Ÿ”ด Investigate now

Something serious: a critical vulnerability with a known exploit, a malware detection, or a high-severity Defender alert.

The rules behind the colours are fixed and published, not a black box. A vulnerability with a CVSS score of 9 or above, a high-severity Defender alert, or an antivirus detection pushes a device to red. Three or more configuration gaps, or a mid-severity finding, makes it amber.

What it looks at

One run asks sixteen questions about each device. You don’t write any of them.

๐Ÿฉบ

Is the device healthy?

Operating system and build, sensor health, onboarding status, join type, exposure level, whether it is internet-facing, and who is signed in.

๐Ÿ›

What is unpatched?

The most severe CVEs affecting it, ranked by score, with the ones that have a known exploit in the wild pushed to the top.

๐Ÿšจ

What has Defender flagged?

Open alerts by severity and category, plus antivirus and SmartScreen detections.

๐Ÿ›ก๏ธ

What has Defender stopped?

ASR rule blocks and warnings, App Control (WDAC) events, Device Control blocks, and Network Protection blocks — the protections that are actively earning their keep.

๐Ÿ‘ฃ

Is anything odd happening?

Living-off-the-land binaries such as certutil and rundll32, outbound traffic on unusual ports, failed logon storms, and executables written into Temp or Downloads.

๐Ÿ“Œ

Is something trying to stay?

Registry Run keys, new services, scheduled tasks, WMI event subscriptions, driver loads and RDP sessions — the classic persistence footholds.

โš™๏ธ

What is misconfigured?

Non-compliant secure-configuration items from Defender Vulnerability Management, grouped by category, each with a fix.

๐Ÿ•ณ๏ธ

What could it not check?

Any section that failed or returned nothing is listed explicitly, so a silent gap never gets mistaken for a clean result.

๐Ÿ“…

Over what period?

You choose the lookback window, from one to thirty days. Seven is the default and is usually the right call.

How you use it

  1. Open Tools → Machine Investigator and sign in. The sign-in is shared with the Advanced Hunting page, so if you are already signed in there you are done. Commercial, GCC, GCC-High and DoD clouds are all supported.
  2. Add your devices. Type a name and press Enter, or click Import List and point it at a .txt or .csv file. Duplicates are skipped and the list caps at twenty-five.
  3. Pick a lookback window. One to thirty days. Shorter is faster.
  4. Choose an engine. Local Rules is the default and the recommendation. Azure OpenAI is there if you want narrative reasoning and have AI Analysis connected.
  5. Click Run Investigation. Progress is shown query by query. All twenty-five devices are batched into roughly a dozen API calls rather than three hundred, so a full run is quick.
  6. Read the verdicts, then open the Details tab on anything that isn’t green.
You do not need to be a local administrator. The results reflect what your Defender account is permitted to see in the tenant, not what rights you hold on your own workstation.

Drilling into a finding

The report is the summary. The Details tab is where you work. It holds four sortable tables:

TableWhat it shows
VulnerabilitiesCVE, score, severity, the affected software and version, and whether an exploit exists. Sorted so anything exploitable is at the top.
EnforcementWhich Defender component acted — ASR, antivirus, App Control, Device Control, Network Protection — what it did, how many times, and the first thing it saw.
Configuration GapsThe secure-configuration item, its category, and when it was last evaluated.
ActivityProcesses, network connections, logons, file writes and persistence attempts, with counts.

Every row has a fix attached

Click any row and a panel opens on the right with a priority badge, a plain description of why it matters, and remediation in three forms — a PowerShell command you can run now, the Intune path to click through, and the Group Policy setting. Vulnerability rows link to the CVE record; enforcement and configuration rows link to the relevant Microsoft documentation.

Configuration gaps come with a verify script. Vulnerability Management data can lag by a day or two, so many secure-config findings include a one-line PowerShell check you can run to confirm the gap is real before you go and fix something that was fixed last week.

Getting it out of the app

If your organisation has set DisableExportButtons, these are blocked and the app says so rather than failing quietly.

Local rules or AI?

โœ… Local Rules Default

  • Runs entirely on your machine. No data is sent anywhere.
  • Deterministic — the same input always produces the same verdict.
  • Fully explainable. Every statement traces back to a specific row and a published threshold.
  • Instant, and free.
  • Works on air-gapped or confidentiality-sensitive investigations.

๐Ÿค– Azure OpenAI Optional

  • Reads the same snapshot and writes the same seven sections in natural language.
  • Better at joining dots across findings and explaining them to a non-specialist.
  • Sends the collected snapshot to your Azure OpenAI deployment.
  • Requires AI Analysis to be connected first, and is throttled to three devices at a time to respect rate limits.
If in doubt, use Local Rules. The AI engine is a convenience for writing up findings, not a second opinion on whether they are real. Both engines see identical data.

For administrators

TopicDetail
AvailabilityPreview feature. Set ShowPreviewFeatures to 1 under HKLM\SOFTWARE\Policies\MDE-Toolkit to reveal it.
Page visibilityVisibilityMachineInvestigatorPage0 hidden, 1 admin only (default), 2 standard users, 3 both.
Licensing / accessA Defender for Endpoint tenant with Advanced Hunting. Results are scoped to the signed-in user’s permissions and device groups.
Data handlingWith the local engine, the snapshot stays in memory on the investigator’s workstation and is written out only if they export it.
Custom guidanceRemediation content lives in JSON catalogs. Point AnalyzerCatalogPath at your own copies, or drop files into %ProgramData%\MDE Toolkit\AnalyzerCatalogs\, to replace the shipped advice with your organisation’s runbooks.
Batch limits25 devices per run; roughly a dozen Advanced Hunting calls total.
Deploy it with Group Policy. ShowPreviewFeatures and every page-visibility value are covered by the downloadable ADMX templates, so you can control who sees Machine Investigator without touching the registry by hand.

Try it

Machine Investigator ships in the current build. Enable preview features and it appears under Tools.

Download MDE Toolkit See the ASR updates