Contacting IT before shutting down a device matters because a shutdown or restart wipes the evidence that answers the two questions that actually decide your response: what malware is running, and where it is talking to. Volatile memory — running processes, network connections, decrypted keys, command-and-control addresses — survives only while the machine is powered on. Powering it off is a deliberate act of evidence destruction that may also leave disk encryption locked. Call your IT or security contact first, describe what you saw, and follow their instruction on whether to isolate, leave it running, or shut it down.

Most people who suspect a breach do the instinctive thing: reach for the power button. It feels like quarantining the problem. In practice, it is often the worst first move you can make on a machine that is still behaving normally enough to keep running. The correct sequence is call, describe, then act on instruction — and the reason is technical, not bureaucratic.

What Actually Disappears When a Device Loses Power

Running memory holds the live state of an attack that a powered-off machine simply cannot reproduce. Every modern operating system keeps process tables, loaded modules, open network sockets, and often in-memory credentials in RAM while it runs. Pull the power and that entire picture is gone within seconds. The hard drive tells you what was installed; it does not tell you what was executing at the moment you noticed something wrong.

Consider a realistic case. An employee opens a document and notices a browser tab they did not open, then a file that will not delete. The machine is still on. At that moment, a forensic examiner could capture the process list, identify the malicious process name, trace its parent process, and see the IP address or domain it is beaconing to. That last item is frequently the single most valuable piece of intelligence in a small-business incident, because it lets the security team block the outbound channel across the whole network rather than cleaning one endpoint. Restart the machine, and the beacon simply re-establishes from a new process with no memory of where it went.

The loss is not gradual. Volatile data degrades the moment the OS shuts down cleanly (processes are terminated, sockets closed, and the kill sequence may overwrite regions), and it is annihilated instantly if someone holds the power button or yanks the cable. The comfortable assumption that "the drive still has everything" is the exact mistake that makes the difference between a contained incident and an open-ended investigation.

There is also a less obvious loss: a powered-off laptop whose disk is encrypted by BitLocker or an equivalent full-disk encryption tool presents a locked volume to anyone who picks it up later. The recovery key may be escrowed in the organization's directory, or it may not. A shutdown taken "to be safe" can quietly convert a recoverable device into a brick until someone locates the key — a delay measured in hours and, in small organizations, sometimes days.

Why IT Wants the Machine Alive, and What They Collect First

Security teams collect volatile evidence in a defined order because the data that disappears fastest is collected first. The standard sequence, established in digital forensics practice and reflected in NIST's guidance on volatile data collection, runs roughly from the least persistent to the most: registers and cache, then routing tables and ARP cache, then running processes, then open network connections, then loaded kernel modules and drivers, then the disk. The order matters because collecting one item can alter another — querying certain process states can change what a later query returns.

A first responder running a live-response collection typically captures the process list with parent-child relationships, the list of established and listening network connections with the associated process IDs, loaded services and scheduled tasks, logged-in users, and the contents of key registry or launch locations. From those they can usually determine three things within the first hour: whether the device is actively communicating with an external address, whether the malicious activity started from a specific user action, and whether other machines show the same process or connection. That third finding is the one that tells you whether you are looking at one compromised laptop or an intrusion touching a dozen.

The practical consequence for the person who called it in is that IT may ask you to do something that feels passive — leave the machine on, leave it plugged in for power, do not browse, do not run a scanner, and do not log in to anything from it. Those instructions exist to keep the volatile state intact for the collection. Running your own antivirus scan, in particular, can overwrite exactly the memory regions the responders need, and it may also trigger the malware to change behavior if it detects the scan.

It is worth naming the counterposition honestly. If a machine is actively encrypting files across a shared drive, or is being used to attack other systems, the irreversibility of that damage can outweigh the value of the evidence, and an authorized shutdown or network isolation is the right call. The judgment belongs to whoever is coordinating the response, which is why the call comes first.

The Call Itself: A Four-Minute Script That Preserves the Evidence

Pick up the phone or open your IT contact channel before touching the device, and give a description structured so the responder can decide in seconds. The script below is not a ceremony; each line supplies information that changes the response decision.

1: Say what you were doing when you noticed it. "I opened an email attachment at 9:40, and at 9:45 the fan spun up and a command window flashed on screen." A user action followed by anomalous behavior points to a specific infection vector. 2: Describe exactly what you see, without diagnosis. Read the pop-up text aloud, name the file and folder that will not delete, describe the window that opened. Do not say "I think I have a virus" — say what the screen shows. 3: State whether the machine is on, connected to the network, and still working. "It's on, it's on Wi-Fi, and it's responding but slow" tells the responder whether live collection is still possible and whether isolation is urgent. 4: Ask one question: do you want me to leave it on, disconnect the network cable, or shut it down? Then do exactly that and nothing else. Write down the instruction and the time you received it.

If you cannot reach anyone quickly, the default posture is to leave the device powered on, disconnect it from the network if the responder is unreachable and you are worried about outbound traffic (on a laptop, disconnect Wi-Fi and unplug the Ethernet; on a desktop, unplug the cable), and keep it plugged into power so it does not sleep or shut down from a depleted battery. Do not keep using it. Do not log in to email, cloud storage, or banking from it. Do not move files off it with a USB stick.

The failure-handling matters as much as the steps. If the machine shuts itself down or restarts before help arrives, note the exact time and whether it was a crash, a Windows update reboot, or something that looked deliberate. That single detail lets the responder distinguish a hardware fault from malware that forces a reboot to complete an installation. If the device blue-screens repeatedly, say so — repeated crashes on a machine that was working yesterday are a recognized pattern, and responders will treat the disk differently when they know it happened.

Shutdown, Sleep, Hibernate, and Restart Do Very Different Things

The distinction between these four actions is not pedantry; each leaves a different trace, and one of them is far less destructive than the others. Restarting and shutting down terminate running processes and clear RAM. Sleeping suspends the machine but keeps memory contents powered, so volatile evidence survives and can still be captured when the machine wakes. Hibernating writes the contents of memory to disk — which preserves some volatile data in a file, but the live process and socket state is no longer running, and the captured image may be encrypted or partially overwritten.

Action Running processes and network connections Volatile memory contents Encrypted disk state Practical effect on investigation
Leave running Intact and live Fully intact Unlocked, mounted Full live response possible
Network isolation only Intact, but connections may time out Intact Unlocked Full live response, loses outbound traffic in real time
Sleep Suspended, not running Preserved in RAM Usually unlocked on resume Partial capture possible after wake
Hibernate Terminated Written to a disk file Varies; file may be encrypted Disk-based analysis only, live state gone
Restart or shut down Terminated Cleared Locked on next boot if encryption enabled Disk-based analysis only, plus key-recovery delay
Power button hold or cable pull Terminated instantly Lost with no clean shutdown Locked on next boot Worst case; risks filesystem corruption too

The table explains why "just restart it and see if the problem goes away" is a bad instruction on a suspected-compromise machine. The restart may well make the visible symptom vanish, because the malicious process has not yet persisted or re-launched — which is precisely the scenario in which you have destroyed the evidence that would have identified it. A machine that behaves normally after a reboot is not cleared; it is unexamined.

Sleep deserves a specific mention because people reach for it without thinking. Closing a laptop lid commonly triggers sleep, not shutdown, and that keeps the volatile state alive in a low-power state. If the responder asks you to preserve the machine, closing the lid is usually acceptable, but confirm it — on some configurations closing the lid triggers hibernate or a shutdown, and a policy setting can change this behavior. Check what your machine does before you rely on it.

When Shutting Down Is the Right Call

Shutdown is the correct move in a narrow set of circumstances, and pretending otherwise would be dishonest advice. The clearest is active, spreading damage: ransomware visibly encrypting files on a mapped network drive, or a machine being used to send phishing messages to your entire contact list. In those cases the harm is happening in real time and every minute counts. Network isolation is the preferred first move because it stops the spread while preserving memory, but if isolation is not possible quickly, an authorized shutdown is defensible.

The second case is physical danger. A device that is overheating, smoking, or making unusual mechanical noise should be powered off and unplugged regardless of what the investigation would prefer. Preservation of evidence never outranks safety.

The third case is a device that will leave the organization's control — a departing employee's laptop that is about to be taken home, or a machine about to be shipped. Again, the responder should be the one to decide, but a locked and powered-off device is far preferable to an unlocked one walking out the door.

Outside those cases, the rule of thumb is that the person who noticed the problem does not decide alone. The cost of a wrong shutdown is asymmetric: shutting down a machine that did not need it can cost hours of investigative work and, in cases where the disk is encrypted and the key is not escrowed, genuine loss of access. Leaving a machine running for twenty more minutes while a responder is contacted almost never makes things materially worse.

Current ransomware groups have made this sharper. Well-documented double-extortion operations copy data out before encrypting, which means that by the time the encryption is visible, exfiltration may already have happened. In that scenario, live memory on the first-identified machine can reveal the tool used to stage and upload the data, and the address it was sent to — evidence you cannot recover from an encrypted disk. The National Institute of Standards and Technology publishes guidance on integrating cybersecurity into the incident response lifecycle; its central point is that identifying and containing an incident depends on accurate, timely information, and the live endpoint is often the only place that information exists.

Where a Breach Is Actually Confirmed: Disk, Logs, and Network

A single device rarely confirms a breach on its own, and the call to IT is partly about connecting the endpoint to the wider picture. The evidence that confirms scope lives in three places: the disk (installed programs, persistence mechanisms, file timestamps), central logs (authentication events, endpoint detection telemetry, firewall records), and network flow data (which internal hosts talked to which external addresses, and when). Shutting down the device does not destroy the last two, which is why a shutdown is not fatal to every investigation — but it does remove the fastest route to correlating them.

Authentication logs are the clearest example. If the same account logged in from an unusual location at 02:14 and then again from the office at 08:30, the logs show both. What they may not show is which process on the endpoint made the connection. That link comes from live network connection data paired with the process ID. When the endpoint is shut down, investigators fall back to firewall logs and endpoint telemetry, which may or may not retain process-level detail depending on what was deployed and how long the retention window is.

This is where small organizations feel the gap most. A company without an endpoint detection product, without centralized logging, and without a designated security contact often has only the live device as a source of truth. That makes the "call before you shut down" rule more important, not less, for exactly the businesses least equipped to follow it.

Preserving a Suspicious Machine When IT Cannot Respond Immediately

There are realistic situations where no one answers for hours: a solo operator, a weekend, a small firm with an outside IT provider on a service contract. You still need a holding pattern that keeps the evidence alive without requiring expertise. The steps are simple and can be written on a sticky note.

1: Disconnect from the network — Wi-Fi off, Ethernet cable out — to stop outbound traffic while keeping the machine running. 2: Keep it plugged into power and disable sleep if you can do so without logging in and browsing. If not, leave the lid open. 3: Do not log in, do not open files, do not run any cleanup or antivirus tool, and do not attach any USB device. 4: Write a timeline on paper: when you noticed it, what you saw, what you clicked before that, what you disconnected and when. 5: Photograph the screen if a message or window is visible — a photograph preserves the exact text even if the window later closes.

The written timeline is not busywork. Investigators routinely reconstruct the user's actions from interviews because endpoint data rarely records them, and a same-day note is far more reliable than a recollection formed two days later. The photograph matters because malware windows often close themselves or change on reboot.

If the machine is a shared or public computer, the same holding pattern applies but the urgency rises, because other people may log in and overwrite the volatile state with their own activity. In that case, contact whoever manages the machine and keep others off it. Locking the screen without logging out preserves the running session while preventing further use — a useful middle step that many people miss.

The Cost of the Wrong First Move, in Hours and Dollars

The difference between calling first and shutting down first is measurable in response time, and response time in a breach has a price. The widely cited IBM/Ponemon "Cost of a Data Breach Report" has for years put the average time to identify and contain a breach at well over 200 days, with organizations that contain incidents faster consistently reporting lower total costs. The report's consistent finding is that speed of containment correlates with cost, not that any single action is the cause. The live-endpoint decision is one lever among many, but it sits at the very start of the clock.

A worked estimate makes the asymmetry concrete. Suppose a small business has one compromised laptop, no endpoint detection, and an outside IT provider who can attend in four hours. If the user shuts the machine down and the disk is encrypted with a key escrowed in a directory the provider can access, the provider loses perhaps one hour reconstructing the timeline and another two collecting disk evidence and re-establishing what was running — call it three hours of billable time that live collection would have compressed. If the key is not escrowed, add a key-recovery process measured in days and, in some cases, a loss of data that cannot be recovered at all. Those figures are illustrative arithmetic, not a published benchmark, but the structure holds: the shutdown converts cheap, fast, live evidence into slower, more expensive, less complete disk evidence.

The second cost is scope. Every hour a beacon keeps running after a shutdown-and-restart is an hour of outbound traffic that may not be logged, and every hour the machine stays off is an hour during which other compromised hosts keep operating undetected. Responders consistently report that the fastest path to understanding scope is the live connection data from the first identified endpoint. Closing that window early is the one decision most likely to turn a one-day incident into a week-long one.

There is a communication cost too. Users who shut down a machine and then report it often describe the problem in retrospect — "it was slow and then I restarted it" — which strips the responder of the details that would have pointed to a cause. The four-minute script above exists to prevent precisely that loss.

For anyone who wants to go deeper on the defensive side, the Center for Internet Security's CIS Critical Security Controls include a specific control on audit log management and another on incident response, both of which exist because endpoint visibility is only as good as the data retained; the controls are freely published and worth reading before you need them.

Building the Rule Into Your Workplace So Nobody Faces This Alone

The rule only works if people know it before the screen misbehaves, because under stress people default to instinct. Three things make the difference. First, a single, obvious reporting channel — one phone number or one chat channel — with an after-hours path, printed on a card taped to the monitor or pinned in the chat tool. Second, a one-page instruction that says, in plain words, "if you think your computer is compromised, leave it on, disconnect the network, and call this number." Third, a named person who is authorized to make the shutdown call, so that the decision has an owner rather than a committee.

Training deserves a specific format. A ten-minute walkthrough at onboarding, repeated annually, in which someone demonstrates a live-response collection on a test machine, is more convincing than a slide deck. When employees see the process list appear on screen from a running laptop, the instruction "leave it on" stops sounding arbitrary and starts sounding like the prelude to something useful.

Tabletop exercises are the other cheap investment. Take a scenario — a phishing email opened, a file server slowing down — and walk through who calls whom, who decides on isolation, and who decides on shutdown. The exercise exposes the gap between the written policy and the actual contact list, which is usually where real incidents stall.

Finally, tell people what to do if they already shut it down. Shame keeps people quiet, and a delayed report is worse than a botched one. The message should be: report it anyway, tell us you restarted it, and give us the time. The disk and the logs are still there, and the investigation continues from a different starting point. Silence is the only unrecoverable error.

FAQ

Does shutting down a computer really destroy evidence if the files are still on the hard drive?

Yes, for a specific and important class of evidence. The disk retains installed files, persistence settings, and some logs, but it does not retain what was running, which processes were talking to which external addresses, or the in-memory keys an attacker was using. Those exist only in RAM while the machine is powered on, and a shutdown clears them.

Is putting a suspected machine to sleep the same as shutting it down?

No, and the difference is significant. Sleep keeps memory powered and preserves the running state, so a responder can often still capture live data when the machine wakes. Hibernate writes memory to disk, and shutdown clears it entirely — both move you from live analysis to disk-only analysis.

What if IT doesn't answer and the machine is clearly doing damage right now?

Disconnect it from the network first, which stops outbound traffic and spread while keeping memory intact. If the damage is active file encryption on a shared drive and isolation is not possible, an authorized shutdown is defensible. Record the time and exactly what you did so the responder can reconstruct the sequence.

Can I just run an antivirus scan to see what's wrong before calling?

Don't. A scan writes to disk, may overwrite the memory regions responders need, and can prompt the malware to change behavior or delete itself. Let the responder decide whether and when to scan, and tell them you were tempted to.

Why does an encrypted laptop that's been shut down cause extra delay?

Full-disk encryption locks the volume at power-off, so the investigator needs the recovery key before reading anything. Keys are often escrowed in an organization's directory, but when they are not, recovery can take days and sometimes fails — a delay entirely separate from the loss of volatile memory.

Do small businesses without a security team still benefit from this rule?

More than large ones, in practice. Without endpoint detection or centralized logging, the live device is often the only place the evidence exists. A one-page instruction and a single after-hours number cost little and preserve the one source of truth a small firm actually has.

How long should a machine be left running while waiting for help?

As long as it takes the responder to arrive or give instructions, provided it stays plugged in and nobody uses it. Keep it on power so it does not sleep or die from a drained battery, and lock the screen rather than logging out if others might otherwise use it.

This article is general educational information about incident response practice, not a diagnosis of any specific device or a substitute for professional security advice. If you are facing a suspected breach, contact a qualified local security professional or your organization's IT provider and follow their direction; if there is immediate danger to people, contact local emergency services.