When the network you manage is also the network you manage it through, an attacker who reaches one device's management interface can reach them all, and an outage or compromise can lock you out at exactly the moment you need control. An out-of-band (OOB) management network is a separate path for administering routers, switches, firewalls, VPN concentrators, hypervisors and server baseboard management controllers (BMCs such as iLO and iDRAC), so that management traffic never shares the production data path. US and allied agencies now say this plainly: the December 2024 joint guidance from CISA, NSA, FBI and Five Eyes partners recommends an out-of-band management network physically separate from the operational data network. This guide covers the architecture: physical versus logical separation, console servers, the administrative access path, BMC isolation, logging and resilience, and how to migrate. Per-device hardening (AAA, SSH settings, SNMPv3, service removal) is covered in the Network Device Management Plane Hardening Guide and is not repeated here.
Key takeaways
- Separate the path, not just the protocol. Encrypted SSH over the production network still lets anyone who can reach the data plane attempt management; an OOB network removes that reachability.
- Accept management only from the OOB network, and block device-to-device management. The 2024 joint guidance asks for both, so that one compromised device cannot be used to log in to its neighbours.
- Never expose management interfaces to the internet. CISA BOD 23-02 requires federal agencies to remove internet-exposed management interfaces, or put them behind a separate zero trust policy enforcement point, within 14 days of discovery.
- Treat BMCs as the most sensitive interfaces you own. CISA's IPMI alert notes that root on a BMC gives control of hardware, software and firmware below the operating system; keep UDP 623 and BMC web interfaces on a trusted management segment only.
- Design the OOB network to survive a production compromise. Independent connectivity, its own authentication fallback and off-device logs are what let you rebuild when the production network is untrusted.
What out-of-band management means
Three terms are often confused. In-band management uses the same interfaces and paths as user traffic. Logically separated management uses dedicated management interfaces, VLANs or VRFs carried over shared infrastructure. Physical out-of-band management uses separate cabling, switches and often separate WAN links that carry nothing but management traffic, plus serial console access that works even when a device has no IP configuration.
| Model | How it works | Protects against | Does not protect against | Fits |
|---|---|---|---|---|
| In-band | Management protocols on production interfaces, filtered by ACL | Casual access, if ACLs are tight | Anyone on the data path probing management; lock-out during outages | Very small sites only, as an interim state |
| Logical OOB (management VLAN or VRF) | Dedicated management interface or VRF, separate routing table, ACLs at the boundary | Reachability from user and server networks | Failures or compromise of the shared switches and links | Branches, mid-size estates, cloud and virtual appliances |
| Physical OOB | Separate switches, cabling and WAN (often cellular or a separate carrier), console servers on serial ports | Production compromise, misconfiguration lock-out, outage of the data network | Compromise of the OOB network itself, or of admin workstations | Data centres, core and edge, providers, critical infrastructure |
| Zero trust mediated access | Policy enforcement point separate from the interface brokers every session | Direct exposure, unauthenticated reachability | Loss of the broker (needs break-glass) | Distributed estates, remote administrators; combines with either OOB model |
Decision guidance: physical OOB is worth its cost wherever losing management means losing the business or being unable to recover, which usually means the data centre, internet edge and core. Logical OOB is acceptable for branches if the management VRF has no route to user networks and is reached only through the administrative access path. A small organisation with a single site can get most of the benefit from a dedicated management VLAN, a single hardened jump host and a console server with a cellular link for the firewall and core switch.
Reference architecture
The design below combines the 2024 joint guidance, CISA BOD 23-02 and the UK NCSC secure system administration principles.
- Administrator devices: privileged access workstations used only for administration. NCSC's first principle is to gain trust in management devices, because an attacker on an admin device can abuse legitimate functions. See the Privileged Access Workstations Guide.
- Administrative access layer: a VPN for administrators, a zero trust access broker, or an administration proxy (jump host or bastion) where authentication, MFA, session control and logging are enforced. NCSC notes that a proxy is a central point for control and logging but is also a single point of failure that needs high availability and break-glass access.
- Management core: dedicated switches or a management VRF that connects the access layer to device management ports, AAA servers (TACACS+ or RADIUS), syslog collectors, NTP, configuration backup and monitoring systems.
- Device edge: management Ethernet ports, BMC ports and serial consoles. Serial consoles connect to console servers, which themselves sit on the management core.
- Independent connectivity: at least one path to each critical site that does not depend on the production WAN, typically cellular or a separate carrier, terminating on the console server or a small OOB router.
Rules that make the architecture work
- One way in: devices accept management connections only from the management core, with VTY and web management ACLs that permit only the access layer and management tools. The joint guidance asks for VTY ACLs to block inbound lateral movement.
- No lateral management: block outbound management protocols (SSH, Telnet, HTTP or HTTPS management) from devices, and monitor for configuration changes that re-enable them. A compromised router should not be able to SSH to the next router.
- No production routes into management: the management VRF or network has no route from user, server or guest networks; production hosts cannot reach management addresses even if ACLs fail.
- Default deny with logging: the joint guidance recommends default-deny ACLs inbound and outbound with denied traffic logged.
- No internet management: management only from trusted devices on trusted networks; internet-facing management ports are removed, not merely filtered.
Console servers and serial access
Serial console is the access method of last resort: it works when the device has lost its IP configuration, when routing is broken, and during boot, which is when firmware and image integrity checks happen. That makes the console server one of the most privileged systems in the estate.
- Authentication: integrate console server logins with central AAA and MFA, and keep a local break-glass account with a vaulted credential that raises an alert when used, as NCSC advises for emergency access.
- Per-port authorisation: grant engineers access only to the ports for devices they administer, not to every port on the console server.
- Session logging: record console sessions and send logs off the console server, because console access bypasses device-level AAA on many platforms when the device is in recovery mode.
- Cellular and dial-in paths: restrict them to VPN termination with certificate authentication; do not leave them open to any caller.
- Lifecycle: console servers are network devices too; patch them, inventory them and include them in configuration backup and integrity monitoring.
Baseboard management controllers
BMCs (Integrated Lights-Out, iDRAC and similar, using IPMI and increasingly Redfish) control power, firmware, virtual media and remote console for servers. CISA's IPMI alert lists the risks: hardware-level control that bypasses the operating system, passwords stored in clear text by IPMI, one password potentially exposing every system in a managed group, outdated network services and few tools to detect a compromised BMC. BMC firmware is a live target: the CISA KEV catalog includes CVE-2024-54085 in AMI MegaRAC SPx, added on 25 June 2025.
| Risk | OOB design response |
|---|---|
| Internet-exposed BMC | BMC ports only on the management network; BOD 23-02 explicitly lists out-of-band server management interfaces such as iLO and iDRAC as in scope |
| IPMI on UDP 623 | Restrict UDP 623 to a trusted management VLAN, scan for it elsewhere, disable cipher 0 and anonymous logins (CISA) |
| Shared or default passwords | Unique credentials per BMC, held in a vault and rotated; central directory authentication where supported |
| BMC to BMC movement | Private VLAN or ACLs so BMCs cannot talk to each other, only to the management tools |
| End of life | Sanitise BMC credentials per manufacturer guidance, or destroy storage where none exists (CISA) |
Shared services on the management network
- AAA: TACACS+ or RADIUS servers reachable only from the management core, with a local fallback account per device for when AAA is unavailable. The Salt Typhoon intrusions show why this matters: Cisco Talos reported the actor changing AAA and TACACS+ server addresses and capturing TACACS and RADIUS traffic to obtain shared secrets. See the Salt Typhoon Telecom Intrusions analysis.
- Logging: the joint guidance asks for logs offloaded from devices, copies stored where they cannot be altered or deleted, encrypted transport (IPsec, TLS) and AAA logs sent to a central server. Keep collectors on the management network and replicate them out of it.
- Configuration backup: store configurations centrally rather than treating the device as the source of truth, and monitor for unauthorised changes. Pull backups over SSH or SCP; Talos observed configurations exfiltrated over TFTP and FTP, so block those protocols from the management network to anything except the backup system, or remove them.
- Time: NTP for all devices from management-network sources, so logs from different devices correlate during an investigation.
- Monitoring: SNMPv3 with authentication and encryption only, with ACLs, as the joint guidance recommends.
Where identity fits
An OOB network concentrates non-human credentials: SNMPv3 users, TACACS+ and RADIUS shared secrets, backup system SSH keys, monitoring API tokens, automation service accounts used by configuration tools, and BMC accounts. Inventory them, store them in a secrets manager, scope automation accounts to the device groups and commands they need, and rotate shared secrets when staff leave or when any device on the network is suspected compromised. Alert on new local accounts and new SSH authorised keys on devices, which Talos lists among Salt Typhoon's persistence methods. See the Secrets Management Guide.
Migration plan
- Inventory management interfaces: every device, its management IP, the interface it is reachable on, the protocols enabled and who can reach it. Scan from user and internet vantage points to find exposure.
- Fix internet exposure first: remove any management interface reachable from the internet, following the BOD 23-02 approach: isolate it or place it behind a separate policy enforcement point.
- Build the management core: management VRF or switches, access layer, AAA, logging, NTP and backup systems, with default-deny ACLs.
- Migrate in rings: start with lab and non-critical devices; move management to the dedicated interface or VRF while keeping the old in-band path as fallback; confirm access, logging and backup through the new path.
- Cut over and remove the old path: apply VTY and web ACLs that permit only the management core, then disable management on production interfaces. Keep console access ready during each change window as the rollback path.
- Add console servers and independent connectivity for core, edge and data centre devices.
- Isolate BMCs: move all BMC ports to a dedicated BMC segment, then rotate their credentials.
- Test recovery: run a game day where the production network is assumed hostile and administrators must reach and reconfigure core devices only through OOB.
Rollback: each device migration keeps the previous management path until the new one is proven; the console port is the final fallback. Record pre-change configurations in the backup system before every cut-over.
Detection and measurement
- Management access from outside the management core: any SSH, HTTPS or SNMP session to a device management address from a source outside the access layer is a policy violation or an intrusion.
- Device-originated management sessions: outbound SSH or Telnet from a network device to another device (lateral management) should be alerted.
- AAA and configuration changes: changed AAA server addresses, new local accounts, new SSH keys, ACL changes and SNMP community changes on any device.
- Break-glass use: every use of a local or console break-glass account raises an alert and a ticket.
- Exposure scanning: scheduled external and internal scans for management ports and UDP 623 outside the management network.
Metrics:
- Percentage of network devices and BMCs whose management is reachable only from the management network (target 100 percent).
- Number of management interfaces reachable from the internet (target zero).
- Percentage of core, edge and data centre devices with working console access, tested in the last quarter.
- Percentage of devices sending logs to the central collector in the last 24 hours.
- Time to regain administrative control of core devices in the recovery exercise.
- Number of shared secrets and BMC credentials older than the rotation policy.
Common failure patterns
- A management VLAN that routes everywhere: the VLAN exists but its gateway sits on the core router with routes to user networks, so separation is cosmetic.
- Jump hosts with direct internet browsing or email: the access layer is only as trusted as the devices that use it; NCSC's management-device principle exists for this reason.
- Exposed edge device web interfaces: internet-facing VPN and firewall web components are repeatedly exploited, and management portals reachable the same way share that risk. The Ivanti Connect Secure Exploitation analysis shows how edge appliance compromise exposes the credentials and service accounts behind it.
- Forgotten BMCs: server teams cable iLO or iDRAC ports into the production network for convenience, with default or shared passwords.
- OOB that depends on production: console servers reached only over the production WAN, AAA with no local fallback, or logging collectors on the production network, so the OOB network fails exactly when it is needed.
- Cleartext management protocols: TFTP, FTP, Telnet and SNMPv2c left enabled because "it is the management network", which is how configurations and secrets are harvested once an attacker reaches it.
Framework mapping
| Framework | Reference | Relevance |
|---|---|---|
| NIST SP 800-53 Rev. 5 | SC-7(15) Networked Privileged Accesses; SC-7(13) Isolation of Security Tools, Mechanisms, and Support Components; SC-37 Out-of-band Channels; AC-17 Remote Access; MA-4 Nonlocal Maintenance; AU-9(2) Store on Separate Physical Systems or Components | Routing privileged access through dedicated paths, isolating management tools, remote administration and protected logs |
| NIST CSF 2.0 | PR.IR-01; PR.AA-05; PR.PS-04; DE.CM-01 | Network protection, least privilege for admin access, logging, monitoring |
| CIS Controls v8.1 | Control 12 Network Infrastructure Management | Secure management of network infrastructure |
| CISA | BOD 23-02 (13 June 2023) | Remove internet-exposed management interfaces or mediate access through a separate policy enforcement point |
| MITRE ATT&CK v20 | T1133 External Remote Services; T1078 Valid Accounts; T1556.004 Network Device Authentication; T1602.002 Network Device Configuration Dump; T1021.004 SSH | Techniques the architecture constrains |
Practitioner checklist
- Scan for and remove every management interface reachable from the internet.
- Build a management network or VRF with no routes from user, server or guest networks.
- Permit device management only from the administrative access layer, using VTY and web ACLs.
- Block outbound management protocols from devices to stop lateral device-to-device logins.
- Place all BMCs on a dedicated segment, restrict UDP 623, disable cipher 0 and use unique vaulted credentials.
- Deploy console servers with central AAA, MFA, per-port authorisation and session logging for core and edge devices.
- Provide independent connectivity to critical sites that does not rely on the production WAN.
- Keep AAA, logging, NTP and configuration backup on the management network with tested local fallbacks.
- Remove TFTP, FTP, Telnet and SNMPv1 and v2c from the management network.
- Run a recovery exercise at least yearly that assumes the production network is hostile.
Standards and references
- CISA, NSA, FBI and partners: Enhanced Visibility and Hardening Guidance for Communications Infrastructure
- CISA: BOD 23-02 Mitigating the Risk from Internet-Exposed Management Interfaces
- CISA: Risks of Using the Intelligent Platform Management Interface (IPMI)
- NCSC: Secure System Administration
- NCSC: Protect Your Administration Interfaces
- NIST: SP 800-53 Rev. 5
- CISA: Known Exploited Vulnerabilities Catalog
- MITRE ATT&CK: Network Device Authentication
Related NHI Mgmt Group resources: Network Device Management Plane Hardening Guide · Edge Device Security Guide · Privileged Access Workstations Guide · Network Segmentation Guide