GTPDoor v3 Hides Its Commands Inside Mobile Subscriber Traffic, While SilentAir Uses Hacked Operators in Six Asian Countries to Redirect Calls, SMS and 2FA Codes

Telecom researchers say the LIMINAL PANDA-linked GTPDoor implant now tunnels C2 through GTP-U on 5–10 or more operators, and a separate state-level actor, SilentAir, abuses SS7 and Diameter to track and intercept mainly Taiwanese subscribers.
Overview
Telecom security researchers have documented a third version of GTPDoor, a Linux implant planted inside mobile operator core networks. It now hides its command traffic in GTP-U, the high-volume protocol that carries subscribers' actual data between operators. The same report describes SilentAir, a separate state-level actor that has used compromised operator nodes in at least six Asian countries to locate subscribers and redirect their calls and SMS. Its main target is Taiwan, and in some cases one-time 2FA codes arrived immediately after interception.
The research was published on 11 November 2026, with an update on 12 November. The researchers have directly seen or strongly suspect 5–10 infected operators and think several dozen may be affected worldwide.
The Campaign: GTPDoor v3
GTPDoor was first analysed publicly in February 2024. Earlier versions hid commands in GTP-C, the control-plane protocol operators use on roaming (GRX/IPX) links. Version 3, which the researchers call GTP-U-Door, moves to the user plane:
- Hiding place. The command payload sits deep inside GTP-U packets, after the tunnel endpoint ID (TEID), mixed in with the normal flow of subscriber traffic.
- Why GTP-U. GTP-U volumes on IPX links are far higher than GTP-C, its layered protocols are harder to inspect, and operators have spent less on securing it. It also carries over into 5G.
- Stealthier listener. v3 captures packets with libpcap instead of a raw socket, which the researchers say keeps it out of
netstatandlsofoutput. - Encoding. Commands are XOR-encoded. The default key matches v1/v2 but can be changed. v3 drops v2's access-control-list feature, and its command set is closer to v1.
- Role. The researchers think v3 is a fallback channel, used when other access is blocked.
Observed commands include starting and stopping exfiltration tunnels through a modified sgsnemu, relaying commands for lateral movement through a component called EchoBackdoor, launching processes including SSH, health checks, and anti-forensics such as file deletion and timestomping. Half of the tunnels lasted under 21 minutes, and a few ran for more than 24 hours.
The activity is tied to LIMINAL PANDA, a cluster other researchers have linked to China-nexus operations with low confidence.
The Campaign: SilentAir
SilentAir is a separate actor with its own tradecraft. The researchers rate it as state-level with high confidence and China-nexus with low confidence. They do not confirm a link to LIMINAL PANDA.
- Infrastructure. Real compromised signalling nodes inside operators in at least six Asian countries, not spoofed addresses.
- Harvesting. SS7 SRI-SM, RestoreData and SendAuthenticationInfo; Diameter AIR.
- Tracking. SS7 AnyTimeInterrogation and Diameter Insert-Subscriber-Data.
- Interception. SS7 UpdateLocation and Diameter Update-Location-Request redirect a target's SMS and calls to a node SilentAir controls.
- Silent SMS. Large volumes, mostly to Taiwanese subscribers around midnight, sent through intermediate SMS centres in Southeast Asia or Taiwan to hide where they came from.
- Targets. Mainly Taiwan, plus about a dozen other countries and territories, including political and government VIPs. Chinese networks and subscribers were not targeted. Bursts ranged from a few subscribers to several thousand in a month.
- Working hours. Activity follows a UTC+8 working day and pauses on weekends and Chinese holidays.
Timeline
All times are Mauritius time (UTC+4).
- February 2024 – first public analysis of GTPDoor (GTP-C).
- November 2024 – LIMINAL PANDA publicly profiled.
- Early 2025 – SilentAir signalling activity starts to appear.
- Q1 2026 – GTP firewalls at several operators start blocking GTPDoor on GTP-C.
- July 2026 – an operator in Luxembourg publicly reports GTP-C GTPDoor activity.
- 11–12 November 2026 – GTPDoor v3 and SilentAir research published.
Indicators of Compromise
- GTPDoor v3 (MD5): 35f0d4d0c5519a2957338e448fce53bc
- SCTPE / SIGweaver SS7–Diameter injection trojan (MD5): da35d98c9a22e8ede93595d21fe7e63e
No IP addresses, domains or Global Titles were published.
Detection
- Network first. Host checks are weakened by the libpcap design, so look at the traffic. Inspect GTP-U for malformed or non-IP T-PDUs, payload after the TEID that doesn't parse as an inner IP packet, and GTP-U to or from nodes that have no reason to carry user traffic.
- Correlate signalling. Flag UpdateLocation or Update-Location-Request messages for subscribers who haven't roamed, AnyTimeInterrogation from unexpected partners, and bursts of silent SMS.
- On Linux core nodes, list processes holding packet-capture handles instead of trusting
netstat:
# Processes with AF_PACKET sockets (packet capture) – review anything unexpected
sudo ss -0 -p
sudo grep -l packet /proc/*/net/packet 2>/dev/null; sudo cat /proc/net/packet
# Unexpected sgsnemu binaries or recently timestomped files
sudo find / -xdev -name 'sgsnemu*' 2>/dev/null
Recommended Actions
- Operators: deploy both GTP-C and GTP-U firewalls and make them share intelligence with SS7 and Diameter firewalls. Treat the IPX interface as untrusted, and monitor internal signalling for correctness, not just the network edge.
- Bring signalling security into the main security operations function, and make sure outsourced core contracts allow forensic access.
- Organisations: stop relying on SMS for high-value 2FA. Use passkeys, FIDO2 keys or authenticator apps, especially for government staff, journalists and executives with links to Taiwan.
Analyst Assessment
Moving the backdoor into the user plane is a deliberate response to defenders getting better at GTP-C. It goes after the protocol operators inspect least and that 5G keeps. SilentAir is the more immediate risk for ordinary users: it shows that SMS one-time codes can be intercepted at the network level without touching the victim's phone. Expect more disclosures as other operators review their IPX traffic. For anyone in the targeted region, moving off SMS-based authentication is the single most useful step.
