Chinese language hackers took management of a goal group’s authentication stack and maintained persistence for 10 years, with full visibility into the executive exercise.
Dubbed “Operation Highland,” the intrusion is attributed to the Velvet Ant cyberespionage risk group, which focused susceptible internet-facing programs earlier than pivoting to a community with no direct exterior path.
Chinese language hackers of the “Velvet Ant” exercise cluster breached the remoted crucial infrastructure community of a big group and performed cyber-espionage operations for 10 years.
The marketing campaign, dubbed “Operation Highland” by Sygnia researchers who found it, started in 2016, concentrating on susceptible internet-facing programs earlier than pivoting to an “air-gapped” atmosphere with no direct web connection.
Velvet Ant’s prolonged espionage operations have been documented in 2024, when Sygnia warned of a marketing campaign concentrating on F5 BIG-IP units that operated undetected for 3 years.
Additionally in 2024, Cisco warned of a zero-day in NX-OS working on Nexus switches, which was exploited by Velvet Ant to realize entry to targets.
Velvet Ant assault chain
The assault begins with the compromise of internet-facing servers, although the researchers don’t point out the particular product or any vulnerability used.
Velvet Ant deployed a modified GS-Netcat reverse shell disguised as a official system element that linked to a hardcoded relay area, offering encrypted distant shell entry.
The shell achieved persistence both through a malicious systemd service or via startup script modification.

Supply: Sygnia
Subsequent, Velvet Ant put in a customized SOCKS5 proxy for community visitors tunneling, enabling it to achieve inside programs that aren’t immediately accessible from the web.
The proxy ran as a daemon masquerading as ‘smbd -D,’ utilizing totally different filenames and ports on every host, and turning compromised servers into inside pivot factors.

Supply: Sygnia
Probably the most fascinating a part of the assault was constructing a distant execution path into the remoted community.
To attain this, Velvet Ant modified the configuration of a compromised internet-facing Nginx server to proxy specifically crafted requests to a compromised backend server.
The backend server’s Nginx configuration was additionally altered to ahead requests to a FastCGI course of (fcgiwrap) listening on a separate port.
The FastCGI wrapper acted as an execution bridge, processing requests and launching a customized binary named ‘uptime.’
The instrument established SSH connections to programs throughout the remoted crucial infrastructure community utilizing parameters equipped in HTTP POST requests.
“By chaining these modifications, Velvet Ant established a remote-execution path into the segregated atmosphere through easy HTTP requests, with no direct connection to the crucial infrastructure community ever required.” – Sygnia
Having established their entry into the remoted atmosphere, Velvet Ant shifted focus to long-term persistence and credential theft by concentrating on Linux Pluggable Authentication Modules (PAM), a set of libraries that allow directors arrange strategies to authenticate customers.
The attackers changed official ‘pam_unix.so’ modules with backdoored variations that settle for hardcoded passwords and harvest consumer credentials.
Sygnia recognized 9 distinct variants of the malicious PAM module, every compiled in a separate construct atmosphere, indicating a well-resourced risk actor.
The researchers say that two of the malicious PAM modules stand out for performing as a backdoor solely and for gathering credentials.
Velvet Ant actors additionally changed OpenSSH elements resembling ssh, sshd, and scp with trojanized variations that captured credentials, logged instructions entered throughout SSH periods, and saved the collected knowledge domestically for future retrieval.
Sygnia says that by extending management to the authentication course of by modifying the PAM and OpenSSH elements, the risk actor had entry to credentials as they have been used within the goal atmosphere and will bypass the authentication circulation.
“Administrative exercise turned totally observable: each login; each command executed throughout compromised hosts. Entry was now not tied to a particular foothold however embedded into the authentication course of itself,” the researchers clarify.
This fashion, the hackers ensured their persistence regardless of password adjustments and session terminations, and lowered “the effectiveness of typical containment measures.”
Complicated cleanup
Sygnia says even after discovering the compromise, remediating it and eradicating Velvet Ant from the compromised atmosphere was notably sophisticated.
The risk actors had changed so many crucial elements with customized variations that eradicating them was more likely to break authentication, lock official directors out, and trigger operational outages.
To deal with this drawback, the researchers constructed a testing lab to validate the binary substitute course of, profiled every host, examined the outcomes, and ready rollback procedures earlier than trying the cleanup.
Sygnia recommends that defenders deal with authentication elements resembling PAM, OpenSSH, and Home windows LSASS as crucial safety property and shield them with EDR, file integrity monitoring, hardened privileged entry, multi-factor authentication (MFA), and steady monitoring for unauthorized modifications.
Organizations ought to plan for offline restoration, which incorporates strict backups with an satisfactory schedule for robotically creating snapshots with immutable copies.
The restoration course of ought to think about testing the backups and restoration hosts working working programs which have been validated, together with the restoration scripts.
Safety groups log 54% of profitable assaults and alert on simply 14%. The remaining transfer via your atmosphere unseen.
The Picus whitepaper exhibits how breach and assault simulation assessments your SIEM and EDR guidelines so threats cease slipping by detection.


