Hackers ran 41 PowerShell modules on a compromised server at a major African stock exchange in just three minutes and 12 seconds.
The toolkit searched for passwords, cloud keys, and database credentials, then tested whether it could access the exchange’s investment application programming interface (API) and a securities depository gateway.
The exchange was not named, and investigators could not determine whether any of the credentials worked. The activity took place on September 20, 2026.
A web shell, a script that lets someone issue commands on a compromised server, ran the modules through Microsoft Internet Information Services (IIS).
Earlier reporting on a suspected AI-written PowerShell backdoor documented a different use of AI-assisted scripting: maintaining remote access. At the exchange, the scripts focused on finding and testing credentials across several systems.
Researchers from Symantec identified the toolkit and detailed it in their November 19 report. They found repeated code patterns consistent with AI assistance but could not identify a model or prove how the scripts were written.
They also found six implants designed to maintain access on the same server, a separate finding that complicated efforts to establish who ran the toolkit.
The modules searched shared policy folders and directories belonging to six business applications. They looked for Amazon Web Services (AWS) keys and database connection strings, which can contain the details an application needs to sign in to a database.
Other modules mapped SharePoint sites, software management servers, and source code infrastructure. A separate case involving AI-assisted Active Directory reconnaissance showed how a PowerShell script could map a Windows network.
Here, the toolkit also reached into cloud and financial systems. Four modules focused on financial interfaces. The operators tried a guessed password, reused a bearer token, and tested API keys against the securities depository gateway.
A bearer token can grant access to whoever presents it while it remains valid. The operators also checked a specific investor number.
Symantec said the choice of targets could point to financial motives, but the investor lookup and interest in internal information meant espionage could not be ruled out.
.webp)
All 41 modules used the same basic script structure, error handling, and code for sending results through Domain Name System (DNS) queries.
That consistency was one reason researchers suspected AI assistance: a developer building a single toolkit might put the repeated code in a shared function instead.
Coverage of generative AI malware development puts the finding in context, but neither repeated code nor rapid execution proves a model wrote the scripts. The pace shows automation, not authorship.
The modules sent encoded output through DNS lookups to a public security testing service. Previous reporting on DNS malware delivery techniques examined a different use of the protocol: delivering malware rather than sending script output.
In the exchange case, Symantec could not see what the remote collector received. Researchers found no evidence of data theft or privilege escalation, but that visibility gap meant they could not determine whether the credential tests succeeded.
What the Witchetty Link Means for Defenders
Symantec attributed the six persistence implants to Witchetty, a China-linked espionage group, with high confidence. It did not attribute the PowerShell toolkit to the group.
The earliest evidence of malicious activity on the server dated to April 2026, though access may have begun earlier. Symantec had previously reported that Witchetty targeted the same exchange in 2022.
That history makes the overlap worth investigating, but it does not establish who ran the September scripts or how they gained access.
For security teams, the clearest signal is the combination of events: a web server worker process launching PowerShell, unusual DNS queries soon afterward, and attempts to access sensitive APIs.
Investigators can compare endpoint, DNS, and authentication logs to see whether those events occurred together and whether any tested credentials were accepted.
A matching file hash can help identify a known sample, but finding a Witchetty implant does not, on its own, connect that host to the 41-module toolkit.
Indicators of compromise (IoCs):-
| Type | Indicator | Source description |
|---|---|---|
| SHA-256 | 8940de2ada979ad2c53ece0b5cb84100c4df0a024b4b363e96cb6a67469e30d5 |
Mimikatz |
| SHA-256 | 5c4f6e21877dfd71e3e95254c6193457e9aab10aca7459ddaeb6d711997190de |
Loader |
| SHA-256 | 982750f4f9edd2ef51b9972a1018a8777bd9d65e0d2e43977a522bdc73ddef5f |
Persistence implant associated with Avast wsc_proxy |
| SHA-256 | 1363534d205ad60f6700d81f2af2c0ce6d959b73fd405adbb598072395bd26e5 |
ShellcodeLoader persistence implant |
| SHA-256 | 0ad7e39ce6b9b7637a002c352637089f6ccf8ae39206a7d895266b21ca26ff0f |
RC4-encrypted loader |
| SHA-256 | 2e223e3c6dc7a130f0281a6c608cd58c153ec026c020c8a661b60a1e285140bf |
Utility associated with beacon.bin and Default.xml |
| SHA-256 | 5a2f093f7ae25c5e381647549496acf375e18dd95263b921518e340ee78e5377 |
DLL loader associated with svchost.exe and WindowsUpdate.xml |
| SHA-256 | b1155f47910f032217352c3a302c01de4d15cabda34331d2866f5d7a52b3ecde |
Loader associated with svchost.exe |
| SHA-256 | b737eae170534f5965b90f4af807cb3f728498f3bccc9355a0afc0171e413230 |
Loader associated with svchost.exe |
| SHA-256 | 5fd9a352d1bec4a11134ee140aa278d07d720b45493a7f01a548293f412972ce |
Loader associated with svchost.exe |
| SHA-256 | 242c411cd415d55e67aae75759e5c07720fcdddc0378b5e35d4c1e4ab2b01eb8 |
Loader associated with svchost.exe |
| SHA-256 | af3737ba032b8c5d3403064899ab6cc48610a397d9f67b626f99b2f0f5082202 |
Loader associated with VSPerfCmd.exe |
| SHA-256 | 6b3061d6e042744bb484b725031e7916b31ad224318a8c0441b9ead14ab853d5 |
DLL loader associated with setting.dat |
| SHA-256 | 9b4dadaa55479727f7c0034ff9c07165597b9a9a9376b6e564634eeaf1c6b876 |
DLL loader associated with setting.dat |
| SHA-256 | d732660db5af253aacbc580d6114d59d233c48b3a5ead6dcfcb55246252abed2 |
DLL loader associated with setting.dat |
| SHA-256 | 85e75e78a545d2fe11ac19385cc8c0789e1056a2cd4edfec32a6a6b113001e5a |
Loader associated with mfeann.log |
| SHA-256 | 1c1704d789332c4d6ce57139d27194ac64c27bfbb1368f1006ffc3c96f3bcb80 |
Nim-compiled DLL impersonating VSPMsg.dll; loader for Msg.log |
| SHA-256 | 3ca9d65046420592e726776bb26b08719721d108f6947fd22af1d5e7cac33edc |
Cobalt Strike beacon |
| SHA-256 | 30dc80cb4c791d80eb21597232fb2f5a4111bf3dfad32c736b789d77127310ef |
Cobalt Strike beacon |
| SHA-256 | 152c125eacdacc69406b120afd95994eef0f7a0cd638c2d8723e64f79344a336 |
Cobalt Strike beacon |
| SHA-256 | 8129e103509f00f40a6a2740b76bc3f9449ac09cd9222ba289e52e86f31f616e |
Cobalt Strike beacon |
| SHA-256 | 26dc22c99e5e88363818e56dc5883258f048e426c9aa515887bf5d06872782f9 |
Not available for analysis |
| SHA-256 | 53ddb6c51aca4bd6b00ed83a8019ee1440351f8eeb95a7b6c74a2b75210d85b2 |
Not available for analysis |
| SHA-256 | af295b5081ccdfd601a0aaecaa81fe1aa92125b548063863dbca2011cef658ce |
Not available for analysis |
| SHA-256 | 0f2c40d31887418de304ca182afb4f4fecfe95937068f11fc798778c2ca95e3f |
mfeann.exe |
| SHA-256 | 002ce0d28ec990aadbbc89df457189de37d8adaadc9c084b78eb7be9a9820c81 |
vsperfcmd.exe |
| Domain | danoam334iqjo0ap2scg6ofosjxfgqhsa[.]oast[.]pro |
DNS destination associated with toolkit output |
Note: IP addresses and domains are intentionally Qdefanged (e.g., [.]) to prevent accidental resolution or hyperlinking. Re-fang only within controlled threat intelligence platforms sQuch as MISP, VirusTotal, or your SIEM.
Give your SOC the alert context on the latest threats to stop incidents quickly -> Integrate Threat Intel from 16K SOCs