Skip to content

Response, New York. Noah Park: Check Point Patches Actively Exploited Management Server Path Traversal. Detection, Dallas. Ava Okello: Report Finds Nearly Half of Deployed Detections Need Attention Before SOCs Can Trust Them.

SOCtember

Always First. Fast SOC News.

Screenshot of the Symantec security.com post titled Major African Stock Exchange Targeted in AI-Assisted Attack, published November 19, 2026 under Threat Intelligence, by Thibaut Passilly and Brigid O Gorman.
Image: Symantec (fair use). The Symantec Threat Hunter Team report published on Thursday.

Threat Intel

Symantec Says 41 Likely AI-Written PowerShell Modules Swept an African Stock Exchange for Credentials in Three Minutes, Tested Them Against Its Investment API and Leaked Results Through DNS

The toolkit ran under a web shell on an IIS server that already carried six implants Symantec ties to the China-linked group Witchetty. Every module shared the same scaffold and the same copy-pasted DNS exfiltration loop to a public out-of-band testing service, which is both the sign of machine authorship and the thing defenders can hunt.

Priya Shah, Threat IntelSingapore6 min read

SINGAPORE - Intruders at a stock exchange in Africa ran 41 PowerShell modules in a little over three minutes to hunt for passwords, cloud keys and API tokens and to test them against the exchange's own investment systems, and the code bears the hallmarks of having been written by an AI model, Symantec researchers said on Thursday.

In a report by its Threat Hunter Team, Symantec, a division of Broadcom, said the modules ran under a web shell on an Internet Information Services server on Sept. 20, between 08:44 and 08:47 UTC, a window of three minutes and twelve seconds, with a new module starting every few seconds. Symantec did not name the exchange or the country.

The researchers called it "one of the clearest examples of suspected machine-written attack tooling that we have seen to date."

Screenshot of the Key findings list in the Symantec report: a toolkit on an African stock exchange with traits of AI authorship, 41 PowerShell modules under a web shell for just over three minutes on September 20, 2026, output sent as hex-encoded DNS lookups, six persistence implants attributed to Witchetty, and earliest evidence from April 2026.
Image: Symantec (fair use).Symantec's key findings.

What the modules went after

According to Symantec, individual modules stole Group Policy Preferences passwords, searched shared folders and the application directories of six business applications for AWS keys and database connection strings, scanned the domain's SYSVOL share and files such as web.config and .env, checked the host for cloud credentials and opened a connection to a database cluster hosted on Amazon Web Services. Others enumerated SharePoint sites and searched their contents for words like "aws," "credential" and "password," queried System Center Configuration Manager and Team Foundation Server, and mapped the IIS sites and application pools on the server.

One module asked SCCM which machine a given user, likely an administrator, had last logged on to, a step Symantec described as lateral-movement targeting.

Four modules targeted the exchange's investment API. They sent a login request with a username and a guessed password, reused a bearer token, and tested several API keys against the gateway of a securities depository, including a blank key that Symantec said may have served as a control. The modules connected to the API host by IP address, set the Host header manually and disabled certificate validation. The attackers also checked a specific investor's number, the researchers said.

Symantec graphic titled Three minutes, twelve seconds: a modular toolkit sweeps an exchange for credentials, showing the window 08:44 to 08:47 UTC and six phases: credential theft, cloud database access, SharePoint discovery, internal service recon, investment API probing and connection handling.
Image: Symantec (fair use).Symantec's breakdown of the toolkit's phases.

Symantec said it did not see the results, so it cannot say whether any of the credentials or keys worked, and found no evidence of privilege escalation or data theft.

Why Symantec thinks a model wrote it

The case for AI authorship rests on uniformity, the researchers wrote. Each module opened with the same error-handling line, used the same outer try and catch block, hex-encoded its output the same way and returned the same empty-output marker. The DNS exfiltration loop was copied byte for byte into all 41 modules. "A programmer building a toolkit would normally write that function once, while a model asked for a standalone script each time has no reason to," the report said.

Variable names were generic and error handling uniformly verbose. One module carried a progress variable no other module had, "as if a model had improved a single script in isolation during iterative prompting," and another submitted an invalid credential as a deliberate test. Symantec said none of the signals is conclusive alone and that it could not identify which model may have been used.

Exfiltration by DNS lookup

Each module split its hex-encoded output into 50-character chunks and resolved a DNS name for each one, under unique subdomains of a single token on oast[.]pro, a public out-of-band testing service of the kind penetration testers use to confirm blind vulnerabilities. The data left as DNS queries rather than web requests and would have been collected by whoever held the token. Symantec called it "an unusual technique."

Screenshot of the Symantec report section DNS exfiltration through a testing service, showing the PowerShell loop that hex-encodes output, splits it into 50-character chunks and resolves each as a subdomain under a token on oast.pro.
Image: Symantec (fair use).The exfiltration loop as Symantec published it.

The subdomain labels also served as a map of the toolkit, with names such as spid1 for host and process enumeration, awslocal1 for cloud and DevOps hunting, gpp1 for Group Policy password theft and ipo1 and cscs1 for queries against investor and market data endpoints, according to the report.

Witchetty, or someone else

The earliest malicious activity on the server dated to April 2026, Symantec said, and six persistence implants there are attributed with high confidence to Witchetty, also known as LookingFrog, a China-linked espionage group first documented by ESET in 2022. Symantec reported in September 2022 that Witchetty had attacked this same exchange. One of the implants was first seen in 2020.

But Symantec did not attribute the toolkit to any group. It said Witchetty may have returned with new tooling, a second and more capable operator may have used Witchetty's access, or an unrelated actor may have compromised the same server, and that it could not rule out any of the three. The focus on cloud secrets and financial APIs differs from earlier Witchetty activity, the researchers noted, and the motive could be financial or espionage.

What SOC teams can hunt

The uniformity that suggests AI authorship also gives defenders something to look for, Symantec said, since the same scaffold and exfiltration loop appeared in every module. For security teams, the practical hooks are:

  • PowerShell spawned by the IIS worker process, w3wp.exe, which is how the modules ran under the web shell.
  • Bursts of DNS queries for long, hex-looking labels, and any resolution of oast[.]pro or similar out-of-band testing domains from servers that have no reason to use them. Symantec published the token domain danoam334iqjo0ap2scg6ofosjxfgqhsa[.]oast[.]pro.
  • Reads of SYSVOL Groups.xml and similar files for cpassword fields, and API calls from a web server to internal financial endpoints by raw IP address.
  • Script block logs showing the repeated scaffold, such as a standalone $ErrorActionPreference = 'SilentlyContinue' header followed by the same hex-encoding routine.

The report lists file hashes for the persistence implants and loaders, a Mimikatz build and four Cobalt Strike beacons found on the server.

About the Author

Priya Shah

Priya Shah covers threat intelligence, intrusion analysis, and adversary tradecraft for SOCtember from Singapore.

Related stories

Threat Intel desk