Hi Khushal, the interesting part here is the
stacking of trust signals: a photo-like filename, a decoy image, an EV-signed first stage, and then a legitimate signed Python interpreter. Each layer looks fairly ordinary in isolation, while the complete chain is plainly malicious. The public ANY.RUN report confirms the sample hash, malicious verdict and Startup persistence indicator.
One wording point is worth tightening: this is
not fully fileless. The padded response and reconstructed CAB may be handled in memory, without saving the CAB itself, but the extracted Python runtime and implant are written beneath
%LOCALAPPDATA%\Wavesoft\..., and
winsec.lnk is written to the Startup folder. “The CAB is reconstructed and extracted from memory” is accurate; “without touching disk” could otherwise give the wrong impression.
A few practical takeaways:
- EV signing does not mean the code is safe. Authenticode establishes integrity and signer provenance, not that the CA inspected the program’s behaviour. Revocation is reactive and does not remove an already-installed payload.
- Renaming genuine
pythonw.exe to winsec.exe does not invalidate its signature because the filename is not part of the signed PE content. The malicious component is the adjacent Python code, especially sitecustomize.py, which Python automatically imports at startup.
- The reported 1/68 VirusTotal result was only a snapshot of recognition at first submission, not a safety score or a useful comparison of antivirus quality.
- Plain WebSocket on port 80 makes this unusually friendly to network investigation. The data is XOR-obfuscated, but it is not protected by TLS, so the captured session can be reconstructed as shown in the full analysis.
For defenders, the strongest hunts are behavioural combinations rather than the sample hash alone:
- execution of an
.scr file with an IMG_<digits> name;
- an executable named
winsec.exe whose internal content matches pythonw.exe;
Lib\site-packages\sitecustomize.py and a lookalike pycached package beside that interpreter;
%APPDATA%\Microsoft\Windows\Start Menu\Programs\Startup\winsec.lnk;
- that interpreter spawning
cmd.exe for dir C: or net config workstation;
- outbound Socket.IO/Engine.IO traffic from a user-profile installation directory.
I would avoid blocking the listed Cloudflare IP addresses globally; they are shared infrastructure. The malicious hostname, HTTP
Host value, file paths, request header and process behaviour are much safer pivots.
Finally, the sandbox
observed host discovery, persistence, profile transmission and two desktop screenshots. That establishes data exposure, but it does not by itself prove password theft or every capability implied by calling it a RAT. If someone ran this outside a sandbox, however, the machine should be isolated and treated as potentially compromised even if a later scan is clean. Preserve the artefacts and seek individualized cleanup through MalwareTips’ Malware Removal Assistance team; if sensitive accounts were used or secrets were visible during the exposure window, protect those accounts from a trusted device as well.
Sources