Symantec researchers say attackers most likely used an AI-written toolkit to hunt for credentials at an African inventory alternate. They ran 41 PowerShell modules in 3 minutes 12 seconds, from 08:44 to 08:47 UTC, underneath a webshell on an IIS server.
The modules examined credentials, a reused bearer token and API keys in opposition to the alternate’s funding API and a securities depository gateway. Symantec couldn’t see the outcomes, so it doesn’t know whether or not any of the credentials labored.
African inventory alternate credential theft
Modules stole Group Coverage Preferences passwords and searched SYSVOL, net roots and utility directories for AWS keys and database connection strings.
Others checked for cloud credentials and opened a connection to an AWS-hosted database cluster. Additionally they enumerated SharePoint, SCCM and TFS servers and mapped IIS websites and utility swimming pools.
4 modules focused the funding API. They despatched a login with a guessed password, reused a bearer token and examined a number of API keys, together with a clean one that will have been a management. They related by IP handle, set the Host header manually and disabled certificates validation.

In response to a report by Symantec, each module opened with the identical error-handling line and the identical strive/catch. Every hex-encoded its output the identical approach and returned the identical empty-output sentinel.
A DNS exfiltration loop was copied byte for byte into all 41 scripts. A programmer would usually write that after as a shared operate.
One module had a progress variable that no different module had. One other submitted an invalid credential as a deliberate check. A brand new module began each few seconds.
Symantec says no single sign is conclusive, however collectively they match AI-assisted authorship higher than bespoke tooling. It couldn’t establish the mannequin.
Every module hex-encoded its output and break up it into 50-character chunks. It then resolved every chunk as a subdomain underneath a single token on oast[.]professional, a public out-of-band testing service.
Information left the server as DNS lookups as a substitute of net requests. Whoever managed the token would have collected it. Defenders can hunt for the shared scaffold and for this loop.
The earliest malicious exercise on the server dates to April 2026, and the attackers might have been current longer. Six persistence implants have been put in. Symantec attributes them with excessive confidence to Witchetty (aka LookingFrog), a China-linked espionage group.
ESET first documented Witchetty in April 2022 as a part of TA410. Symantec reported in September 2022 that Witchetty hit two Center Jap governments and an African inventory alternate, utilizing ProxyShell and ProxyLogon to plant webshells.
Symantec doesn’t attribute the toolkit to any group. It could possibly be Witchetty with new tooling, a extra succesful second operator utilizing the identical entry, or an unrelated actor. This toolkit is much extra superior than something beforehand tied to Witchetty.
The motive is unclear. Probing of the funding API and banking gateway factors to monetary acquire. A test on one particular investor’s quantity, and curiosity in supply code and worker knowledge, level to espionage.
Symantec discovered no proof of privilege escalation or knowledge theft. Symantec Endpoint merchandise block identified malicious recordsdata, and the Symantec Safety Bulletin lists present protections.








