Bitdefender says EggStreme's operators rebuilt the malware 12 and 26 days after its report; cause is inference
Bitdefender's second EggStreme report dates the first rebuilt Fuel build 12 days and the first rebuilt loader 26 days after its first report (derived). The page calls its report a likely trigger; that is an inference, and victims, access and cause are not stated.
By Parminder Kumar Sharma · LinkedIn · · 19 min read
Bitdefender dates the first rebuilt Fuel build 12 days after its report and the first rebuilt loader 26 days after (derived)
On 18 November 2026 Bitdefender published a second report on its EggStreme malware family, "EggStreme Malware: Rebuilt Under Scrutiny". It says the first report went out on 17 September 2025, that the first EggStremeFuel build with a new configuration format was compiled on 29 September 2025, "twelve days later", and that the first loader build in the newer format followed "14 days after that", with a compile timestamp of 2025-10-13 07:35:48. Add the two gaps and the loader build is 26 days after the first report (derived: 12 plus 14, or 13 October minus 17 September). Bitdefender's summary says the operators "revised their malware architecture and infrastructure, and the timing suggests our report was a likely trigger".
That is the checkable fact: three dates on one page, and the arithmetic between them. This is what it does not establish:
- Why the operators changed anything. The reaction to the report is Bitdefender's inference from sequence. The page's own disclaimer calls statements about the operators' motives, timing or reactions "inferences from observed activity". The page does not discuss any other cause, such as another vendor's report or a change that was already under way.
- That all the changes came after the report. The page says some minor agent changes were tracked before the report, and a passive-mode agent build is stamped 2 September 2025, 15 days before the report (derived).
- When the new builds were first used. The 12 day and 26 day figures run from the report to a compile timestamp. For the loader the page says the build was seen in the wild the next day. For the Fuel build it gives no sighting date.
- Who the victims are, how many there are, or how the operators first got in. The page names no organisation, publishes no victim count and describes no first access.
- Who the operators are. The page keeps a "suspected Chinese APT group" attribution and says it is "not a statement about the involvement of any government". This site makes no claim about any government.
What changed, and the dates the page prints for each change
The new page sets out a build timeline as a figure, captioned "The EggStreme build timeline shows which changes came before and after our initial report." The figure is an image, and this site read only the page text, so the timeline below is drawn from dates printed in that text, not copied from the image. Most of the dates are compile timestamps. The page says compile timestamps on some agent builds are earlier than they were first seen, so it dates those by first sighting, not by build date.
Dated changes in the new Bitdefender page, with the days after the first report (derived)
| Date as printed | Days after 17 Sep 2025 (derived) | What the page says |
|---|---|---|
| 2025-09-02 | 15 days before | A passive-mode agent build is stamped this date. The page dates agent builds by first sighting where compile stamps are earlier, and does not say which applies to this one. |
| 2025-09-29 | 12 | First EggStremeFuel build with the new configuration format (AES-256-CBC replaces RC4 keyed from the file name). The page says twelve days after its report. |
| 2025-10-13 07:35:48 | 26 | First loader build in the newer format (RC5-32 CBC replaces RC4). 14 days after the Fuel build. The page says it was seen in the wild the next day, so it treats the stamp as reliable. |
| 2025-10-28 | 41 | Three loader builds with a custom 8-byte block cipher, stamped within about 95 minutes of each other. The page: about six weeks after publication. |
| 2025-11-05 | 49 | A fourth custom-cipher loader build, and a second passive-mode agent build. |
| 2025-11-11 | 55 | An older-format loader build, RC4 with a key tied to the machine. |
| 2025-12-16 and 2026-05-14 | 90 and 239 | Loader builds with a second custom cipher (variant B). |
| Not stated | Not stated | The move away from the single certificate authority. The page says no new domains have appeared under the legacy authority since the first article. |
Two things in the table do not fit a clean "report, then rebuild" story, and the page says both. The older loader format did not disappear: the page says older-format containers are still in use and now carry newer agent builds, and the 11 November loader build above uses the older RC4 scheme. And the page says the framework is built so that capabilities are chosen at compile time from one codebase, which it gives as the reason the changes came fast. A change of cipher, in its words, "does not require a new framework, only a new build".
A likely trigger is an inference: what the sequence can and cannot carry
The page argues the point from the nature of the work. An advanced persistent threat (a state-aligned intrusion team, in the page's gloss) depends on staying unseen, and once its tooling is public defenders can hunt for it. So, the page says, exposed operators change what defenders now know. The sequence on the page fits that argument: the first report, then a changed configuration scheme 12 days later, then a changed loader 14 days after that.
A sequence is not a cause. The same dates would appear if the operators had planned the changes already, and the page's own entries show changes stamped before the report. The page does not claim otherwise, and its summary says "suggests" and "likely". The table below separates what the page states as fact from what it infers, and from what it does not say at all.
What the new Bitdefender page states, infers and leaves unsaid about the reaction to the report
| Question | Stated or inferred | Not stated |
|---|---|---|
| Were the operators reacting to the report? | Inferred: the timing suggests the report was a likely trigger. | Any statement from the operators; any other cause considered, such as another vendor's report or planned work. |
| When was the first Fuel build with the new configuration format made? | Stated: compile timestamp 2025-09-29, twelve days after the report. | When it was first seen in the wild; where it was used. |
| When was the first newer-format loader built? | Stated: compile timestamp 2025-10-13 07:35:48, seen in the wild the next day. | How many hosts it reached. |
| Did every change follow the report? | Stated: some minor agent changes were tracked before it; most major changes, including infrastructure, were first observed after it. | Which minor changes came before; how they were dated. |
| Did the operators leave the old infrastructure? | Inferred: they appear to have moved away from the single certificate authority; the page assesses it has been abandoned or phased out. | The date of the move; whether every campaign has its own authority (the page says it has not confirmed that). |
| Does the older tooling still run? | Stated: older-format containers are still in use and now carry newer agent builds. | How many intrusions use each. |
Two further limits come from the page's own wording. First, it dates some agent builds by first sighting because their compile stamps are earlier, and a compile stamp is written by whoever builds the file, so the page leans on the next-day sighting for the loader. The Fuel stamp has no such cross-check on the page. Second, what a vendor can see is bounded by its own telemetry: a change that predates Bitdefender's first sighting looks to Bitdefender like a change after the report.
The page's final section adds a general point, that exposing an operation can cost its operators more than one campaign because tooling is shared across operations. It then raises a question about handing such operations to autonomous, AI-operated tooling, and says "The data in this article does not test that question." This site treats that as the vendor's own commentary and draws nothing from it.
Rotation against rebuild: which of the changes cost the operators something
Bitdefender frames the response with a model of indicators ranked by what it costs an attacker when defenders detect them: hash values at the bottom, then IP addresses and domain names, then network and host artifacts, then tools, then tactics, techniques and procedures at the top. The page groups the bottom three tiers as "trivial". Its own sentence on what the operators did: they rotated hashes, IPs and domains (trivial); changed payload paths, configuration formats and a firewall rule name (artifacts); and reworked Fuel, the loader and the agent (tools). The behaviours stayed: DLL sideloading, lateral movement through file shares, in-memory execution and mutual TLS. In the page's words, "the operators spent at the tools level and kept their methods."
"Rotation" is the cheap response: new hashes, new domains. "Rebuild" is what the page says it saw as well: a different network protocol and command set in the agent, a new configuration scheme in Fuel, a new cipher and payload layout in the loader, and a passive agent that never beacons out. The page also says the new ciphers add little real protection: "Neither custom variant offers real protection", because every payload file carries its own key. What the swaps defeat, it says, is detection and tooling tied to a specific routine, not analysis. That is a statement about cost to defenders, not about operator skill.
The page separates two kinds of change that are easy to confuse: the step from the framework of the first report (v1) to the newer generation (v2), and the variation between campaigns inside v2. Within a generation, it says, builds share architecture and protocol while cryptography, configuration and capabilities change per campaign. The keylogger is its clearest example: it is missing from every newer agent build the page analyses, which Bitdefender reads as a capability compiled into the earlier build for that target, not a module dropped.
Where the two pages, and the new page itself, differ
This site read both Bitdefender reports in full. The differences below are reported as printed; none is resolved here.
Differences between the first report, the new report and the page itself, as read on 18 November 2026
| Point | What each page prints | How this site handles it |
|---|---|---|
| Date of the first report | The new page: "Our first article went out on September 17, 2025." The first page's byline and page data: 17 September 2025. The first page's indicator file link is named with the prefix 2025_09_10. Other coverage seen while searching gives 10 September; that coverage was not read. | 17 September, the date both pages print. If the date were 10 September, the page's "twelve days" would be 19 days to 29 September (derived), so the page's own figure fits 17 September. |
| How the attribution is worded | The new page: "The attribution is unchanged from our first report: a suspected Chinese APT group." The first page's summary opens "A Chinese APT group compromised a Philippine military company"; its body says the tactics are "consistent with those of Chinese APT groups". The word suspected does not appear on the saved first page. | Reported. The new page calls the attribution unchanged; the first page's summary states it without that qualifier and its body words it as consistency. |
| Who was targeted | The first page: a Philippine military company. The new page: Philippine government organisations with multiple victims at high confidence; Taiwan at low to medium confidence; it "broadens the target description". | Both confidence levels quoted. No organisation is named by either page, and none is printed here. |
| How many loader samples | The page says Bitdefender identified four new loader samples soon after the first article (one RC5, three custom cipher). Its table lists seven newer-format builds, the last stamped 2026-05-14. | Both reported: the later builds are not "soon after" in the page's text. |
| Reaction against schedule | The summary: the timing suggests the report was a likely trigger. The text: some agent changes were tracked before the report; a passive build is stamped 2025-09-02. | The two are kept apart in the timeline. The brief does not say all changes were a reaction. |
| The two generations | The comparison table says almost everything detectable at file, protocol and infrastructure level changed. The text says older-format containers still run and carry newer agents. | Both reported. The table is a summary of trends, not a clean cut. |
The first report said the way the logon script reached the share was unknown. The new report adds no first-access finding. A fuller account of what the first report covered is on Bitdefender's page; this site does not restate it.
Victimology and attribution: what Bitdefender assesses
The new page says: "we assess with high confidence that the operation continues to target Philippine government organizations, with multiple victims". It adds that, based on its analysis of the samples, Taiwan is targeted "at low to medium confidence", and that readers "should treat that finding as a lead, not an established fact". It says "We do not name victim organizations, and we do not publish a victim count." The first report covered a single Philippine military company, so the stated target set is wider than before.
On attribution the page says this update "makes no new attribution claim" and that the reason it treats the newer activity as the same operation is technical continuity. Its evidence is a legacy domain that links the Fuel activity to the loader and agent activity (which it used to track as two clusters), older-format loader containers that carry newer agents, DLL sideloading as the delivery pattern, and one plaintext string used as key and initialisation value in every newer loader variant. The attribution itself stays "a suspected Chinese APT group", with the page's statement that this is a technical assessment and not a statement about the involvement of any government. This site repeats the assessment as Bitdefender's and takes no position on it.
Lateral movement: the same strategy with new mechanics
The page says the overall strategy did not change: stage a payload on storage the targets already reach, then trigger execution remotely. In the first report that meant Impacket-style tooling, the Stowaway proxy and a logon script injected into a compromised share. In the recent case, the page says, "the operators staged on an unmanaged system", a machine outside endpoint management that other workstations could reach. That system hosted VBS script files and a Fuel package of three parts: a legitimate executable, a malicious DLL for it to sideload, and a configuration file. The page does not say the recent case used Impacket.
The page says the operators ran the script remotely over SMB, and that it copied the Fuel package to the target and set persistence through a run key. It says the same method was seen on multiple machines in the same network, and "The pattern suggests that valid credentials" and a reachable share were enough. It adds that the hashes in its indicator tables "will be stale by the next campaign", while the pattern will probably still apply: an unmanaged host serving scripts, SMB-triggered execution and a run key under a familiar-looking vendor path.
Each name was chosen to look routine, the page says. The staged updater imitates OneDrive's own: the genuine file is OneDriveStandaloneUpdater.exe, under the Program Files OneDrive folder on a per-machine install and directly under the user's OneDrive folder on a per-user one, while the staged copy has a shorter name and sits two folders deeper, in setup and logs. The printer administration script prncnfg.vbs is a real Windows script, which is the point of using its name.
What the page prints: names, paths and the comparison, copied
The new page prints artifacts. This section copies the ones that help a defender hunt, exactly as the page prints them and attributed to it. It leaves out the keys, the page's decryption code and the IP addresses, and it prints no hash list: the page publishes its full indicator set separately, in a file it links on a code hosting site, which this site did not read. Domains are defanged as the page defangs them. The page cautions that the current owners of listed domains and addresses may be unrelated to the activity.
The six named components, as the first report introduced them and the new page repeats them:
EggStremeFuel stage one; protects its configuration (RC4 in v1, AES-256-CBC in v2)
EggStremeLoader decrypts the next stage and places it in memory
EggStremeReflectiveLoader loads the agent without a file on disk
EggStremeAgent the backdoor; 58 commands in v1, command IDs 21 to 88 in v2
EggStremeWizard a secondary backdoor
EggStremeKeylogger not seen in any newer agent build the page analysed
The recent lateral movement case. What a defender hunts, as printed in the page's table of observed elements. The run value points at a path two folders below the genuine updater's:
Run value (persistence at logon, logged-on user only):
name: StandaloneUpdater
key: HKEY_CURRENT_USER\Software\Microsoft\Windows\CurrentVersion\Run
data: %LOCALAPPDATA%\Microsoft\OneDrive\setup\logs\StandaloneUpdater.exe
attrib hide and unhide of .tmp files and StandaloneUpdater.exe on a share under Users\Public
Execution wrapper: C:\Users\Public\Officelog.bat
WScript.exe running: C:\Users\Public\prncnfg.vbs
Genuine OneDrive updater, for comparison (the page):
OneDriveStandaloneUpdater.exe under C:\Program Files\Microsoft OneDrive\ (per machine)
or directly under %LOCALAPPDATA%\Microsoft\OneDrive\ (per user)
Where the loader hides its payload. The page lists these payload paths from its own indicator set, each chosen to look like an ordinary Windows data file, and says content that does not match the extension, or a non-system process reading the file at startup, gives them away:
%SYSTEMROOT%\Media\Alarm00.wav newer format
%LOCALAPPDATA%\Microsoft\Feeds\Feeds.tmp newer format
%PROGRAMFILES%\Common Files\Microsoft Shared\msinfo\msinfo32.dll newer format
%SYSTEMROOT%\apppatch\appmain.sdb older format
%SYSTEMROOT%\apppatch\expmain.sdb older format
%SYSTEMROOT%\Globalization\ICU\systemZones.res older format
%LOCALAPPDATA%\ConnectedDevicesPlatform\CDPGlobalSettings.cdpresource older format
Fuel configuration, newer builds (imitates the genuine wpndatabase.db in the same folder):
%LOCALAPPDATA%\Microsoft\Windows\Notifications\wpndatabase.dat
Agent configuration, most newer files recorded (seven distinct hashes at this one path):
%SYSTEMROOT%\system32\config\systemprofile\AppData\Roaming\Microsoft\Crypto\Crypto.db
The loader builds the page tabulates, filename, compile timestamp and scheme as printed (the page also prints an MD5 for each; those are omitted). Read down the timestamps: the newer format appears on 13 October 2025, the older one is still built on 11 November, and every newer-format loader is named after a Windows system library. The page says the key and initialisation value in every newer-format loader variant is the same plaintext string, KERNEL32.
Loader builds as listed in the new Bitdefender page, read 18 November 2026
| File name | Compile timestamp | Scheme |
|---|---|---|
| iscsiexe.dll | 2024-05-15 07:35:43 | RC4, key google (original variant) |
| DevicesPlatform.dll | 2025-06-09 13:54:20 | RC4, key Windows (older format) |
| xwizards.dll | 2025-07-09 08:42:01 | RC4, key svchost (older format) |
| advapi32.dll | 2025-10-13 07:35:48 | RC5-32 CBC, PKCS#7 (newer format) |
| advapi32.dll | 2025-10-28 13:57:10 | Custom CBC, PKCS#7 (newer format) |
| advapi32.dll | 2025-10-28 15:29:11 | Custom CBC, PKCS#7 (newer format) |
| advapi32.dll | 2025-10-28 15:31:16 | Custom CBC, PKCS#7 (newer format) |
| dbghelp.dll | 2025-11-05 00:41:20 | Custom CBC, PKCS#7 (newer format) |
| dwmapi.dll | 2025-11-11 07:23:49 | RC4, host-bound key (older format) |
| advapi32.dll | 2025-12-16 09:03:34 | Custom CBC, PKCS#7 (newer format) |
| advapi32.dll | 2026-05-14 08:33:39 | Custom CBC, PKCS#7, contains debug strings (newer format) |
The passive agent. Two newer agent builds, stamped 2025-09-02 and 2025-11-05, run in passive mode: they open a listening port instead of beaconing out. The page prints the details a host log would show:
Listener: TCP port 5060 (the standard SIP port)
Firewall rule: inbound allow rule named "Core Networking - SIP"
profiles: domain, private and public; edge traversal enabled
created through the COM firewall interfaces INetFwPolicy2 and INetFwRule
Domains the page names as linking the generations, defanged as printed. The page says mepintei[.]com connects the Fuel activity to the loader and agent activity, and flags readyui[.]net and hostkbc[.]com as original-infrastructure domains seen with the newer agent builds. The page's reason for printing infrastructure is that this was observed in connection with the activity; it does not say these resources are owned by the operators today.
mepintei[.]com used by newer Fuel configurations and binaries; also among the agent's original-infrastructure domains
readyui[.]net original-infrastructure domain seen with the newer agent builds
hostkbc[.]com original-infrastructure domain seen with the newer agent builds
ATT&CK technique identifiers the page prints for the recent case and the rebuilt components. The page's full mapping has 27 technique identifiers, including several not repeated here (token impersonation, WMI, discovery, proxying and reflective loading in the agent). It lists these for the lateral movement chain and the persistence and evasion pieces:
T1021.002 SMB/Windows Admin Shares scripts executed on targets over SMB
T1570 Lateral Tool Transfer Fuel package copied from the staging host
T1574.002 DLL Side-Loading loader DLLs named after system libraries
T1547.001 Registry Run Keys / Startup Folder StandaloneUpdater run value
T1059.005 Visual Basic WScript.exe running prncnfg.vbs
T1059.003 Windows Command Shell Officelog.bat
T1036.005 Match Legitimate Resource Name or Location OneDrive styled paths, prncnfg.vbs, Alarm00.wav, wpndatabase.dat
T1564.001 Hidden Files and Directories attrib hide and unhide on staged files
T1078.002 Domain Accounts one set of compromised credentials apparently reused across machines
T1562.004 Disable or Modify System Firewall the Core Networking - SIP rule
What Bitdefender recommends, and what to do in order
Bitdefender's own defender action lines, once, as printed. The page's closing recommendations are headed "Recommendations for defenders, whether or not they run Bitdefender products", and they also name Bitdefender products as its way to implement each layer (its attack surface reduction, process, network, fileless and extended detection offerings and its threat intelligence platform). Bitdefender sells those products. The lines below are the page's, not independent findings.
Scripts and run keys:
Alert on WScript.exe or cmd.exe running scripts from C:\Users\Public\, and on run-key values that point into
%LOCALAPPDATA%\Microsoft\OneDrive\setup\logs\ or any OneDrive subfolder that is not the vendor's installed updater path.
Fuel configuration:
Treat any non-Windows process that reads or writes %LOCALAPPDATA%\Microsoft\Windows\Notifications\wpndatabase.dat as suspicious
(and capture process memory before remediation if a Fuel infection is suspected: the per-run RC4 traffic key exists only there).
Passive agent:
Alert on a firewall rule named "Core Networking - SIP" that is not in your Windows baseline, and on listeners on TCP 5060
on hosts that are not VoIP endpoints.
Constants:
Build at least one detection per constant above, because a rule tied to a constant is likely to outlive the next rebuild,
while a rule tied to a hash from this article probably will not.
Take this with you
This site's suggestions, in order (not Bitdefender's advice)
- Hunt for the page's own items first: scripts running from C:\Users\Public\, run-key values that point into the OneDrive setup logs folder under a user profile, and a StandaloneUpdater.exe two folders deeper than the genuine updater's location.
- Treat DLL side-loading from user writable paths as a detection priority, and look for Windows system library names such as advapi32.dll, dbghelp.dll and dwmapi.dll loaded from outside the system directories.
- Watch writes to SMB admin shares and remote script execution from machines that are not under endpoint management. Inventory those machines: the page's staging host was reachable by workstations and outside endpoint management.
- After any published report on a family you have hunted, hunt again: an actor that rebuilds will not match last year's indicators. Hunt behaviours, which is the page's own Pyramid of Pain point, not hashes.
- Check hosts for new inbound firewall rules and for listeners on TCP 5060 where no telephony runs.
- If you hold government or defence supply chain data in the region, share what you find with your national CERT.
- Read the headline for what it is: the page says the timing suggests the report was a trigger. It does not prove cause, and it is the vendor's own account of its own report.
The question the report leaves open
Bitdefender says the operators rebuilt within weeks, and also that what they kept was how they work: sideloaded DLLs under system library names, staging through file shares, running from memory and requiring mutual TLS. If an actor can rebuild its tooling within a month of being published and keep its behaviour, how much of your defence is hashes and domains from the first report?
Key facts
Sources
- PrimaryEggStreme Malware: Rebuilt Under Scrutiny, the 18 November 2026 report (70 minute read, read in full by this site on 18 November 2026): the build timeline, v1 and v2 comparison, lateral movement, victimology, infrastructure and defender actionsBitdefender Business Insightsaccessed 2026-11-18
- PrimaryEggStreme Malware: Unpacking a New APT Framework Targeting a Philippine Military Company, the first report dated 17 September 2025 (read in full by this site on 18 November 2026): the baseline framework, its attribution wording and its first victim descriptionBitdefenderaccessed 2026-11-18


