Symantec says attacker ran SYSTEM scripts on a domain controller via UPS software's feature; entry not stated
Symantec says an attacker used CyberPower PowerPanel Business, UPS management software, to run scripts as SYSTEM on a Windows domain controller in a five stage chain. The post does not say how the attacker got in, who they are, when it happened or who was hit.
By Parminder Kumar Sharma · LinkedIn · · 15 min read
Five stages, ten script names, 19 hashes, and no stated way in (derived counts)
On 18 November 2026 Symantec's Threat Intelligence and Research Team published a post titled "Behind the Power Outage: Unmasking a New Intrusion Vector in Enterprise Infrastructure Tools" (the date is in the saved page's own header). It describes one intrusion in which an attacker used CyberPower PowerPanel Business (PPB), software that manages UPS devices, "as their execution channel on a Windows domain controller". Counting the post, the account has five stages, names ten script files, and lists 19 distinct SHA-256 values (all derived counts: the arithmetic is in the key facts and the fact check).
What the post establishes is narrow and specific. The product has a documented setting that runs a command file when a power event fires. The attacker put their own scripts where that setting looks, and the product's service ran them as SYSTEM. Symantec says no complicated memory-corruption exploit and no new service or scheduled task was needed for that part. It is not the same as saying no vulnerability was involved anywhere: the post names four possible ways the attacker could have reached the console (three documented weaknesses and stolen administrator credentials), and does not say which one, if any, was used.
Symantec sells security products and wrote the post. It says its endpoint products "will detect and block" files from its list that are malicious and available to it. CyberPower makes PPB, and neither saved page carries a statement from CyberPower. This brief reads one primary (Symantec's post) and one secondary (Infosecurity Magazine, 19 November 2026), and names no author or reporter.
What happened, in the order Symantec gives it
Symantec's post walks through the intrusion in five stages. The diagram follows its headings; the boxes outside the solid chain are what the post does not say.
Stage 1, scripts go in through the console. The domain controller's PPB web service on TCP port 3052 received requests to its file upload endpoint, which "accepts a file". Symantec shows two traces of where an uploaded file lands: a Tomcat temporary file in the product's web server work directory, and then a script scanned at a path showing it passing into the extcmd folder. The post does not say whether those requests were authenticated, and it does not say whether the upload route is the one the advisories it mentions describe.
Stage 2, the service runs them as SYSTEM. Once a script sits in extcmd, the PowerPanel service, ppbed.exe, runs it. Symantec says the commands in the chain have ppbed.exe as their ancestor and run as SYSTEM. It also says scripts in extcmd are named pp followed by five random lowercase letters, so that the names blend in with the product's own files. The post's own tables and indicator list print four script names that do not fit that pattern (see the section on differences).
Stage 3, a general remote shell. Symantec writes that the attacker "did not use their access for triggering actions from power events", and instead used it to run arbitrary commands through cmd.exe. The nine commands in its table do the following (purposes as printed, command lines not reproduced here): test that the domain controller can reach the attacker's server, decode a staged binary, transfer a tool, verify a download, execute a payload, load an assembly reflectively, stage two PowerShell scripts, and list network connections. One downloaded tool, wupd.exe, is detected as MSIL.Killav!gen1, which Symantec calls a .NET antivirus killer. A 4,096-byte .NET file, byte-identical to hello.exe, only prints a greeting; Symantec says it was a test that the in-memory load path works.
Stage 4, the Active Directory database, revised until it worked. Symantec says the attacker developed the collection scripts on the fly: run, check the log, change, run again. It lists nine script names in a table, and a tenth in its indicator list. The table is below.
The Stage 4 scripts as Symantec lists them (purposes shortened; the post's table of 18 November 2026, read from a copy saved on 19 November 2026)
| Script as printed | Symantec's stated purpose, shortened | Tool named |
|---|---|---|
| ppskbcx.cmd | Exports a copy of the Active Directory database, ntds.dit | ntdsutil |
| ppgxrte.cmd | Loads an embedded .NET program into memory | Not named in the row |
| ppgtpui.cmd | Changes to run in the background with output sent to a log file | ntdsutil |
| ifm.cmd | The same ntdsutil export | ntdsutil |
| ifm2.cmd | The same export, plus a log file and a done marker file | ntdsutil |
| ifm3.cmd | The same export, with a timestamp, a tool check and an exit code in the log | ntdsutil |
| ppotuff.cmd | Saves the SYSTEM registry hive, then copies ntds.dit from a shadow copy | esentutl |
| ifm4.cmd and ppinthw.cmd | Copy ntds.dit from a shadow copy | esentutl |
Nine script names across eight rows: five use ntdsutil, three use esentutl and one loads a .NET program (5 + 3 + 1 = 9, derived). The pivot from ntdsutil to esentutl matters, Symantec says, because it lets the attacker copy a locked ntds.dit through the Volume Shadow Copy service. The post also says the cycle of writing, uploading, running and checking "bears the hallmarks of an interactive operation with a human at the keyboard".
Stage 5, an implant from the same channel. After the collection scripts, PowerShell started upnpcont.exe from a directory under the SYSTEM profile. Symantec says that name is associated with a Microsoft UPnP component and does not belong in a SYSTEM profile directory, and that the file loaded an atl.dll of unknown reputation from the same directory, the typical pattern of DLL side-loading. The process lineage the post prints for a later collection task is seven names long: vssadmin.exe, powershell.exe, upnpcont.exe, powershell.exe, cmd.exe, ppbed.exe, services.exe. That is six parent links, and ppbed.exe sits five levels above vssadmin.exe (derived by counting the post's own line). Processes the implant spawns keep ppbed.exe as the root of their lineage.
A feature, not a bug: why a patch is the wrong first question
Symantec's central point is about the product's design, not a defect in it. Its closing section says the command execution feature "is a double-edge sword", and that it "simply accepts a script, stores it, and runs it as SYSTEM through a signed service on a server that often sits in a privileged position". The advice that follows, to treat UPS and power management software like a remote monitoring and management tool, is the post's own.
Why such hosts attract an intruder is also in the post. PPB is usually installed on machines that must shut down cleanly in a power failure, which Symantec lists as hypervisors, database servers, file servers and, here, domain controllers: "These also happen to be the systems that an intruder most wants to control." It then names three properties such machines share, and the diagram sets each beside what a defender could log. The right column is this site's suggestion, not Symantec's.
What this means for patching. Symantec's advice does include "ensure patches are up to date" and the post says public advisories document hard-coded credentials, a default administrator password and a file upload path-traversal flaw in PPB. Any of those, or stolen administrator credentials, "could give an attacker the necessary console access". Patching and changing the default password close those doors. But the post's account of stage 2 describes the product doing what its documentation says, which a patch for a door does not change. This site's reading: if the event action feature is working as designed, only access control, credentials, network limits and monitoring stand between a console login and SYSTEM. The saved pages do not say whether CyberPower has changed, or plans to change, how the feature accepts scripts, and print no CyberPower advisory or statement. This site did not read the advisories Symantec refers to; the post prints no CVE or advisory identifiers for them.
The telemetry label. Symantec says that in process telemetry "the activity looks like a power-management product launching cmd.exe", which defenders may dismiss as a standard power-related recovery process. A power management product is expected to run a script when the power fails. That is the friendly name doing the work of a control: the vendor, the signed service and the product category all read as benign, so a child process of ppbed.exe gets less scrutiny than the same child process of a web server. Symantec's remedy is to check the parent process. It notes that the attacker's tooling was almost entirely built in Windows tools (it lists 7: certutil, ntdsutil, esentutl, reg, vssadmin, robocopy, wmic), so file reputation alone has little to catch.
What the post states, and what it does not
Symantec's post of 18 November 2026 against the questions a reader will ask (read from a copy saved on 19 November 2026)
| Question | Stated by Symantec | Not stated |
|---|---|---|
| How did the attacker reach the console? | Possible routes: three documented weaknesses (hard-coded credentials, a default administrator password, a file upload path traversal flaw; the post says the advisories document issues "including" these) or stolen administrator credentials. | Which route, if any, was used; whether requests were authenticated; any CVE or advisory identifier. |
| Was a vulnerability exploited? | No complicated memory-corruption exploit and no new service or scheduled task was needed. | Whether a flaw was used to get console access. |
| Who did it? | Not attributed to a known actor. | Any group, any country, any motive beyond that they are unclear. |
| Who was hit? | A Windows domain controller running PPB. | The victim, its sector or its size. |
| When? | The post is dated 18 November 2026 in its header. | Any date or time of the intrusion, how long the attacker was present, when the implant started. |
| How many hosts or installs? | One domain controller is described. | Other PPB hosts abused, how many PPB consoles are reachable from the internet, how many installs exist. |
| What happened to the copied database? | ntds.dit copied by script; no exfiltration or final payload such as ransomware seen. | Where the copy went, whether it left the network, whether credentials were used. |
| What does CyberPower say? | Nothing on the saved pages. | Any advisory, fix, or change to the feature; whether CyberPower knows. |
The second row is the one to read twice. "No exploit" in Symantec's sentence is about the execution step. It is not a statement that the attacker needed no flaw to get in, because the post does not say how they got in.
Where the two pages differ
Infosecurity Magazine reported the research a day later. The differences below are in wording, and this site resolves none of them: the article follows Symantec's page.
Symantec's post and Infosecurity's report of it, side by side (both read from copies saved on 19 November 2026)
| Point | Symantec's post | Infosecurity Magazine |
|---|---|---|
| What was not seen | Says "no exfiltration or final payload such as ransomware was seen". | Says "no data theft or ransomware seen". Data theft is not the post's word. |
| Was a bug needed? | Says no complicated memory-corruption exploit and no new service or scheduled task was necessary. | Says this meant "the attacker could run code without a software bug". The post's wording is narrower. |
| Headline verb | Describes abuse of a documented feature of a trusted application. | The headline says PowerPanel was "Exploited". |
| Human operator | Says the cycle bears the hallmarks of a human at the keyboard, and later that it points to live human presence. | Says Symantec read the rewrites as a sign of a human operator working live. |
| Credentials advice | If indicators are found, assume Active Directory credentials are exposed and plan a krbtgt reset and wider credential rotation. | Says it told organisations to reset Active Directory credentials if indicators turned up. |
| Script names | Says extcmd scripts are named pp followed by five random lowercase letters, and lists four names (ifm.cmd to ifm4.cmd) that do not fit. | Does not mention the naming. |
| Dates | Dated 18 November 2026 in its header; no date for the intrusion. | Says the research was published 18 November; no date for the intrusion. |
| Vulnerability identifiers | Prints none. Refers to public advisories. | Prints none. |
The script naming line is a difference inside Symantec's own page. Counting the names it prints in its tables and indicator list gives ten: six fit the pp plus five lowercase letters pattern and four (ifm.cmd, ifm2.cmd, ifm3.cmd and ifm4.cmd) do not (derived by pattern match). A detection that looks only for the pp pattern would miss those four. The post's own audit advice, to compare the extcmd folders with the product defaults, does not depend on the pattern.
What to do, in order
Items 1 to 6 are Symantec's advice in this brief's words: the first five from its recommendations section, in its order, the sixth from its closing section. Items 7 to 9 are this site's suggestions.
Take this with you
Symantec's advice first, then this site's suggestions
- Symantec: locate every PPB console instance and restrict TCP port 3052 to named administrator hosts. Do not expose it to the internet, other servers or user networks.
- Symantec: change the default administrator password and make sure patches are up to date.
- Symantec: audit the extcmd and temp\profile\extcmd paths on every PPB host and compare them with the product defaults. Treat any unknown script as hostile and review the console's event-to-command assignments.
- Symantec: review the techniques and indicators in its post. If you find them, assume Active Directory credentials are exposed and plan a krbtgt reset and wider credential rotation.
- Symantec: treat machines running power management software as high value assets and defend them accordingly.
- Symantec: treat UPS, power management and other infrastructure software like any other potentially dual-use remote management product, with network restrictions, unique credentials, patching and process level monitoring.
- This site's suggestion: inventory PPB across the estate, including domain controllers, hypervisors, database servers and file servers, and record who can reach each console.
- This site's suggestion: alert on cmd.exe, powershell.exe or certutil.exe whose ancestor is the PPB service, ppbed.exe, which is the telemetry the post describes. Do not rely on the pp naming pattern alone.
- This site's suggestion: ask your CyberPower contact in writing whether any advisory or fixed release covers the console's script feature, because no saved page prints a CyberPower position. Ask the separate question whether a domain controller needs a UPS agent with command execution at all.
What Symantec prints: names, paths and 19 hashes
The post prints defensive artifacts, and this section reproduces only those, attributed to it. It contains no script content and no command lines. The post also prints an address and ports for the attacker's staging server, defanged; this site does not reprint them, so use the post for those. Hash values and file names are copied exactly as printed; the post's dash between a hash and its name is replaced by a colon here.
First the names and paths. These are what a defender searches for on a host: the service and its port, the upload endpoint with the host replaced by a placeholder (this site defanged the scheme), the folders scripts pass through, the implant and its DLL, and the script naming rule with its exceptions.
Product: CyberPower PowerPanel Business (PPB)
Service: ppbed.exe (starts under services.exe, runs as S-1-5-18, SYSTEM)
Web console: embedded Tomcat, TCP port 3052, REST API
Upload endpoint: hxxp://[HOST]:3052/local/rest/v1/file/upload
Tomcat work directory: web-server\work\tomcat\localhost\local
Upload temp file: upload_[RANDOM].tmp
Script staging: ...\CyberPower PowerPanel Business\temp\profile\extcmd\
Script run from: ...\extcmd\
Script naming (Symantec): pp?????.cmd (pp and five random lowercase letters)
Other script names listed: ifm.cmd, ifm2.cmd, ifm3.cmd, ifm4.cmd
Implant: C:\Windows\System32\config\systemprofile\AppData\Local\50679f292a1\upnpcont.exe
Side-loaded DLL: atl.dll (same directory)
Windows tools the post names: certutil, ntdsutil, esentutl, reg, vssadmin, robocopy, wmic
Event variables the documentation names: %EVENT%, %MODULE_NO%
Next the process lineages. Both are printed in the post. The second is the one that matters for detection: every process the implant started traced back to the PowerPanel service.
ppbed.exe → cmd.exe → powershell.exe → upnpcont.exe
vssadmin.exe ← powershell.exe ← upnpcont.exe ← powershell.exe ← cmd.exe ← ppbed.exe ← services.exe
Last the hashes. Symantec's list has four headings and 19 rows, and all 19 values are distinct (7 + 6 + 4 + 2 = 19, derived). Two values, the upload temp file and hello.exe, also appear in the post's body, so the page prints 21 times in all. One file name can carry several values: ifm.cmd has three and dt.exe has two, so a name match alone is not an indicator (this site's reading). Symantec says that if an indicator is malicious and the file is available to it, its endpoint products detect and block that file.
File indicators
22b8f2776f1c941d8240ee4243da537147f8d2ee4d859a68038335da7b0d7688: ppskbcx.cmd (ntdsutil ifm)
04206431c83d1d81726e56fffb5166c0a7f138eb366e5aa918dee54af1ef22b5: ppgtpui.cmd (ntdsutil ifm, background with log)
899f072c505e56ed65b4ea5d64ab3bc0f98c39366b2c413c988551843e729451: ppgxrte.cmd (reflective .NET load)
76670eaadda9806c9aefa91c2ee4f2a33dcf8124574721a6ff072f19e7c03514: pppwngl.cmd (reflective .NET load with try/catch)
839300af16b7117adf05751e1847e1e026146d388cc328a179ba799eb1c9c229: ppotuff.cmd (reg save, esentutl /vss)
2776cfcdd46264043b823adaf37e0c85d0efd3b46d5fdbb72219535061d44b86: ppinthw.cmd (esentutl /vss)
1949894c26e03d6f15ed7f4393a89dbd52f8acc62959d3d5e4f695d6a19c6d9c: upload_[RANDOM].tmp (Tomcat upload temp file)
AD collection scripts
cbd790d052a57d18f1cd58fe4c9887340eec5d38bfae105caf0d222f30a856d5: ifm.cmd
c3f6aa4dc9cdbcd2cfdacd18e59e0c1af0406d59d1f81cd962ac3cde60626ee8: ifm.cmd
b0d382443411a7986ce7eda473c3e8ed787de9083db71deb4ea0539ed74c0c25: ifm.cmd
9abbcb4ac9a98690c666a2e843facf62af773d078db40570f08f052b0e42003f: ifm2.cmd
c2454052f49e43226f69413ce58ba0bb3a2d89e7a52f32ef8adbf0961bb838c2: ifm3.cmd
8ee88ef957597fd4788ae9da4463dcbecec0f1969e6573204f4263f99f46b125: ifm4.cmd
Staged tools and loaders
3d73d5342a11dce35081e70a4ea75d39ad65bc6d480337bd33c1d2e0ba4090ee: dt.exe
12e716f48875befa978ff4a78a0b99a18b1809fc5bcc34c761158875da611aa8: dt.exe
bdea608430fbf6384ce518a542af3e0e69a685980ef826f7db3de20d4c835e26: wupd.exe
d8ae46604dc4eab411977764bd923af686aea117fb33304520d928f1d382ef0e: hello.exe
Implant and side-loaded DLL
5b2285f788eec4d17bf18d85795d457f40f7598bbab2697919f5810f54a9c3fe: upnpcont.exe
d4032e61aa9ef01a5a174a90beffa5a27dd21b36e653ce5304907787554b7d95: atl.dll
The question the post leaves open
The post describes a domain controller whose own shutdown helper was told to run attacker scripts as SYSTEM, and it stops before the questions that decide how widely to worry: how the attacker reached the console, when, on how many hosts, and what became of a copy of the directory database. If the product that shuts your domain controller down cleanly will run any script a console login hands it, and your process telemetry would read that as power recovery, which of your detections would ever notice?
Key facts
Sources
- PrimaryBehind the Power Outage: Unmasking a New Intrusion Vector in Enterprise Infrastructure Tools, dated 18 November 2026 in its header: five stage account, three properties, recommendations and indicator list. Read in full from a copy saved 19 November 2026 at about 17:23 UTCSymantec (security.com)accessed 2026-11-19
- Reported byCyberPower PowerPanel Exploited to Run Malicious Code on Domain Controller, 19 November 2026: a report of Symantec's post, used for the differences in wording only. Read in full from a copy saved 19 November 2026 at about 16:31 UTCInfosecurity Magazineaccessed 2026-11-19


