The IP address in your sign-in and firewall logs is rarely the attacker's own: it is a compromised home router, a rented server, a Tor exit or an internal host relaying for them. MITRE ATT&CK tracks this as T1090 Proxy: adversaries direct traffic through intermediaries to avoid direct connections to their infrastructure. In current operations that means residential proxy networks, operational relay box (ORB) networks of compromised edge and IoT devices, commercial VPNs and Tor, and internal port proxies on hosts already compromised. This guide explains the four T1090 sub-techniques in ATT&CK v20, where proxying shows up on the inside and outside of the network, why IP blocklists keep failing, and which detections and identity controls hold up. It is written for SOC analysts, threat intelligence teams, identity engineers and network architects, and stays defensive. For C2 in general, see the Command and Control (C2) Guide.
Key takeaways
- Source IP is weak evidence. Mandiant reported that an IPv4 address may belong to an ORB network for as few as 31 days, and exit nodes often sit on residential ISPs in the target's own country.
- Inside the network, proxies are built from built-in tools. ATT&CK records Volt Typhoon using netsh port proxy and FRP; detect the configuration change, not the traffic alone.
- Identity is the strongest control point. Phishing-resistant MFA, device-bound sessions and conditional access remove the value of a clean-looking residential IP.
- Anonymity infrastructure is still common. The Salesforce data exfiltration campaign used Mullvad VPN and Tor, which tenant IP restrictions can refuse outright.
- Track networks, not indicators. Mandiant advises treating ORB networks as evolving entities, so intelligence should describe infrastructure behaviour and lifetime, not just lists of IPs.
How the attack works
ATT&CK v20 places T1090 in Command and Control on Windows, Linux, macOS, ESXi and network devices. Proxies hide where traffic really comes from or goes to, and reduce the number of external connections a defender can see.
| Sub-technique | What it is | Examples recorded in ATT&CK v20 |
|---|---|---|
| T1090.001 Internal Proxy | A compromised internal system relays C2 between other internal systems and the outside, reducing external connections | Volt Typhoon using netsh port proxy; APT29 using SSH port forwarding during the SolarWinds compromise; Turla using compromised internal systems as relays; tools such as HTRAN |
| T1090.002 External Proxy | An outside intermediary (compromised host, VPS or port redirector) masks the C2 destination | APT29 using compromised residential endpoints as proxies; Quad7 Activity starting SOCKS5 proxies on compromised devices; APT28 using other victims as relays |
| T1090.003 Multi-hop Proxy | Chains of proxies, Tor or chains of compromised network devices, so defenders see only the last hop | Salt Typhoon and Volt Typhoon using multi-hop proxies; SPACEHOP and FLORAHOX activity routing through chains of devices and a custom Tor relay layer; Tor use in the Salesforce data exfiltration campaign |
| T1090.004 Domain Fronting | Different domains in the TLS SNI and the HTTP Host header so a CDN routes traffic to a hidden destination | APT29 using the meek plugin for Tor |
Proxying links to several Resource Development and C2 techniques: T1584.008 Network Devices and T1584.005 Botnet (compromising routers and IoT for relay capacity), T1583.003 Virtual Private Server, and T1665 Hide Infrastructure, which ATT&CK uses for APT29 choosing residential proxies in the same ISP range as the victim and for Quad7 rotating compromised SOHO addresses. Tunnels often carry the proxied traffic; see the T1572 Protocol Tunnelling Defensive Guide.
ORB and residential proxy networks
Mandiant's May 2024 research describes ORB networks used by China-nexus espionage actors as meshes of rented VPS nodes and compromised routers and IoT devices, often end-of-life. It separates provisioned networks built from leased VPS (SPACEHOP, also tracked as ORB3), non-provisioned networks built mainly from compromised devices (FLORAHOX, ORB2, recruited with the FLOWERWATER router implant) and hybrids. Traffic enters at a relay node, crosses traversal nodes and leaves at an exit node near the victim. Criminal residential proxy services work similarly at consumer scale: the FBI warned on 5 June 2025 (I-060525-PSA) that streaming boxes, projectors and other home IoT devices were being enrolled in the BADBOX 2.0 botnet and residential proxy services, sometimes before purchase. See the Operational Relay Box Networks Guide.
Where it shows up
- Identity sign-ins: password spraying and token replay from residential or VPN addresses near the victim's users. ATT&CK records Quad7 Activity routing password spray attacks through chains of compromised network devices.
- Edge device exploitation: Mandiant reports a SPACEHOP node being used to exploit a NetScaler vulnerability (CVE-2022-27518) in late 2022. See the Edge Device Exploitation by Threat Actors Guide.
- SaaS data theft: the ShinyHunters Salesforce data theft campaign ran test queries and full-table extraction from Mullvad VPN and Tor addresses; in the Snowflake breach, logins came mostly from Mullvad and Private Internet Access VPN addresses.
- Inside the network: port proxies on servers, SOCKS proxies on compromised hosts and NAT or port forwarding rules on firewalls relay access between segments.
- Your own devices as someone else's proxy: unpatched routers, cameras and NAS devices you own can be recruited into ORB or residential proxy networks, which is an exposure in its own right.
Impact
- Location-based controls fail: geo-blocking and "unfamiliar country" rules do not fire when the exit node is local.
- Blocklists age fast: short-lived ORB addresses and rotating residential IPs make indicator sharing less useful and raise false positive risk on shared consumer address space.
- Attribution is harder: Mandiant notes an egress IP may now point to an ORB network rather than a specific actor, so attribution relies more on tools and tradecraft.
- Segmentation is bypassed: internal proxies let an intruder reach networks that have no direct internet path.
- Liability: compromised corporate devices used as proxies can carry attacks against other organisations.
Detection
ATT&CK v20's analytics for T1090 look for internal relays (unusual processes such as rundll32 or svchost connecting to internal peers not seen in baselines; socat, ssh, iptables or ncat port forwarding on Linux and ESXi; new NAT or port forwarding rules on network devices), outbound relays to VPS, CDN or anomalous autonomous systems, and SNI to Host header mismatches for domain fronting.
| Signal | Telemetry | Sub-technique | Tuning note |
|---|---|---|---|
netsh interface portproxy rules created, or new listeners on unusual ports on servers | Command-line logging, EDR, registry auditing of PortProxy keys | .001 | Legitimate port proxy use is rare; alert by default |
| Host-to-host connections on SMB, RPC or high ports between machines that never talk | Flow logs, Sysmon 3, EDR network events | .001 | Peer-group baselining reduces noise |
| New NAT, port forwarding or proxy rule on a firewall or router outside a change ticket | Configuration diffs, AAA accounting | .001, .003 | Mirrors the network device analytics in v20 |
| Sign-ins from anonymous or malicious IP addresses, atypical travel, or anomalous token use | Identity provider risk detections (for example Entra ID Protection Anonymous IP address and Malicious IP address) | .002, .003 | Microsoft defines Anonymous IP address as sign-ins from addresses such as Tor or anonymous VPNs; residential proxies are not named, so pair with device signals |
| Same account active from a managed device and an unmanaged session at once | Sign-in logs with device IDs, session logs | .002 | Works even when the proxy IP is clean and local |
| TLS SNI differs from the HTTP Host header | TLS-inspecting proxy logs | .004 | Needs inspection (M1020); many CDNs now block fronting |
| Outbound connections from your routers, cameras or NAS to unknown hosts, or inbound sessions at odd hours | Firewall and flow logs for device segments | Your devices as proxies | Feed into the IoT asset inventory |
Measuring it
- Phishing-resistant coverage: percentage of workforce and admin accounts that can only sign in with phishing-resistant methods.
- Anonymous access refused: share of SaaS tenants with IP restrictions or conditional access blocking Tor and commercial VPN exits for interactive and API sign-ins.
- Internal relay detections: time to detect a test port proxy or SOCKS relay in authorised exercises.
- Edge hygiene: count of end-of-life routers, firewalls and IoT devices still internet-exposed, trending to zero.
- Intelligence freshness: median age of proxy IP indicators in blocklists, with automatic expiry.
Prevention and hardening
ATT&CK v20 lists M1037 Filter Network Traffic (block known anonymity networks and C2 infrastructure, noting fronting can bypass it), M1031 Network Intrusion Prevention and M1020 SSL/TLS Inspection. Because the attacker controls where their traffic appears to come from, the durable controls bind access to identity and device rather than to network location.
- Make IP reputation a signal, not the gate: require phishing-resistant MFA and compliant or registered devices for sensitive access, so a residential exit IP alone gets nothing. See the Phishing-Resistant MFA Guide.
- Refuse anonymity infrastructure where you can: in SaaS and data platforms, set login IP ranges, network policies or conditional access to block Tor and commercial VPN exits; the Salesforce campaign analysis recommends exactly this.
- Restrict non-human identities by network: API keys, service accounts and integration tokens should work only from known egress ranges or workload identities, since they cannot do MFA.
- Prevent internal relays: block netsh portproxy and unapproved proxy tools with application control, restrict server-to-server traffic to required flows (micro-segmentation) and route all outbound web traffic through an authenticating proxy (NIST SP 800-53 SC-7(8)).
- Do not feed ORB networks: patch or retire end-of-life edge and IoT devices, disable remote management on internet-facing interfaces and monitor device configurations. See the Edge Device Security Guide.
- Use TLS inspection selectively: where legally and technically possible, inspect uncategorised and CDN traffic from servers to catch fronting and proxied C2.
Rolling it out
- Run conditional access or tenant IP rules in report-only mode for two weeks and list affected users, partners and integrations.
- Agree exceptions (travelling staff using corporate VPN, partner ranges) and enforce for administrators first, then all staff.
- Deploy internal relay detections in alert mode on servers, then add blocking application control rules by server tier, with the previous policy kept for rollback.
- Add blocklist expiry (for example 30 days) so stale proxy indicators do not accumulate.
Choosing an approach: small organisations gain most from phishing-resistant MFA, device compliance and blocking anonymous IPs at the identity provider. Large organisations should add internal relay analytics, configuration monitoring on network devices and an intelligence process that tracks proxy networks as entities, as Mandiant recommends.
Lessons from incidents
- Clean-looking access from VPN exits was the norm. In both the Snowflake breach and the ShinyHunters Salesforce campaign, data was taken through valid sessions from commercial VPN or Tor addresses. Tenant-level network restrictions and MFA would have refused or flagged those sessions regardless of the credential.
- Network devices are both target and relay. The Salt Typhoon telecom intrusions analysis describes packet capture through chains of devices, and ATT&CK records the group using compromised routers and multi-hop proxies. See the Salt Typhoon Threat Actor Profile and the Volt Typhoon Threat Actor Profile.
Framework mapping
| Framework | Reference | Relevance |
|---|---|---|
| ATT&CK v20 mitigations | M1037 Filter Network Traffic; M1031 Network Intrusion Prevention; M1020 SSL/TLS Inspection | Blocking anonymity networks, malware signatures, finding domain fronting |
| MITRE D3FEND | D3-NTCD Network Traffic Community Deviation; D3-CAA Connection Attempt Analysis; D3-OTF Outbound Traffic Filtering | Hosts talking outside their peer group, relay connection patterns, egress control |
| NIST SP 800-53 Rev. 5 | AC-17 Remote Access; SC-7(3) Access Points; SC-7(8) Route Traffic to Authenticated Proxy Servers; SI-4(11) Analyze Communications Traffic Anomalies | Governed remote access, few controlled egress points, anomaly analysis |
| NIST CSF 2.0 | PR.AA-03; DE.CM-01 | Authenticate users, services and hardware; monitor network services |
Practitioner checklist
- Enforce phishing-resistant MFA and device compliance for administrators and sensitive apps.
- Block Tor and commercial VPN exits at the identity provider and in SaaS tenant network settings.
- Restrict API keys and service accounts to known egress ranges or workload identity.
- Alert on netsh portproxy, SOCKS tools and new listeners on servers.
- Baseline host-to-host flows and alert on new peer relationships.
- Diff firewall and router configurations for new NAT or forwarding rules.
- Retire or patch end-of-life edge and IoT devices so they cannot join ORB networks.
- Expire proxy IP indicators automatically and track proxy networks as entities.
- Test detections for internal and external relays in authorised exercises.
Standards and references
- MITRE ATT&CK: Proxy (T1090)
- MITRE ATT&CK: Multi-hop Proxy (T1090.003)
- MITRE ATT&CK: Hide Infrastructure (T1665)
- Google Cloud (Mandiant): China-nexus cyber espionage actors use ORB networks
- FBI IC3: Home Internet Connected Devices Facilitate Criminal Activity (I-060525-PSA)
- Microsoft Learn: What are risk detections?
- MITRE D3FEND: Network Traffic Community Deviation
- NIST: SP 800-53 Rev. 5
Related NHI Mgmt Group resources: Operational Relay Box Networks Guide · Command and Control (C2) Guide · Proxies and Forward vs Reverse Proxy Security Guide · T1102 Web Service Defensive Guide · MITRE ATT&CK T1078 Valid Accounts Guide