Most serious intrusions into routers, switches and firewalls do not need an exploit: the attacker logs in to a management interface with a credential, a community string or a shared key that was never protected well enough. The management plane is everything an administrator or a management system uses to reach a device: SSH and web consoles, SNMP, AAA traffic to TACACS+ or RADIUS servers, configuration transfer, logging and time. This guide sets out a concrete hardening baseline for that plane with command-level examples, a protocol reference table, a phased rollout and the detections that show it is working. It is written for network engineers, security architects and SOC leads who own enterprise or service provider infrastructure, and it goes deeper on configuration than the broader Edge Device Security Guide.
Key takeaways
- Reachability comes first. A management interface that only accepts connections from a separate out-of-band network and named admin hosts removes most of the attack surface before any password policy matters.
- Device AAA is its own identity system. CISA and its partners advise a central AAA service with phishing-resistant MFA that is not linked to the primary corporate directory, with local accounts kept for emergencies only.
- Plaintext and legacy protocols must go. Telnet, HTTP management, FTP, TFTP, SSH version 1 and SNMP v1/v2c leak credentials and configurations, and the Salt Typhoon telecom intrusions show how attackers harvest them.
- Device secrets are non-human credentials. TACACS+ keys, SNMP strings, local password hashes, NTP and routing keys need an inventory, strong storage types and rotation like any other machine secret.
- Configuration is evidence. Store the authoritative configuration off the device, alert on drift, and treat a changed AAA server address or a new local account as an incident until proven otherwise.
What the management plane is and why attackers target it
Network devices separate traffic into three planes. The data plane forwards user traffic, the control plane runs routing and switching protocols, and the management plane carries administration. Whoever controls the management plane can read and change everything else.
The joint guidance Enhanced Visibility and Hardening Guidance for Communications Infrastructure, published on 3 December 2024 by CISA, NSA, FBI, the Australian Signals Directorate's ACSC, the Canadian Centre for Cyber Security and NCSC-NZ, was written in response to intrusions into telecommunications providers.
The Salt Typhoon telecom intrusions illustrate the pattern. Public reporting described attackers mostly logging in with valid stolen credentials, collecting device configurations over TFTP and FTP that contained SNMP read/write strings, recovering local passwords stored with weak encryption types offline, and capturing TACACS+ and RADIUS shared secrets from network traffic. Once inside, they changed AAA server addresses, created local accounts, altered ACLs and enabled SSH on high ports. Only one case was reported as involving a vulnerability, the Cisco Smart Install flaw CVE-2018-0171. See the Salt Typhoon Threat Actor Profile.
| Attacker goal | Management plane weakness | ATT&CK technique |
|---|---|---|
| Log in as an administrator | Reused or stolen credentials, no MFA, management reachable from user networks | T1078 Valid Accounts; T1021.004 SSH |
| Harvest secrets from configuration | Configs sent over TFTP or FTP, Type 7 or Type 5 password storage, SNMP write access | T1602.002 Network Device Configuration Dump; T1552.001 Credentials In Files |
| Capture AAA and SNMP secrets in transit | TACACS+ obfuscation, SNMP v2c community strings in clear text | T1040 Network Sniffing |
| Keep access | Local accounts created, SSH keys added, AAA server repointed | T1098.004 SSH Authorized Keys; T1556.004 Network Device Authentication |
| Hide or tamper with the device | Unverified images, no off-box logging | T1601 Modify System Image; T1601.001 Patch System Image |
| Reach other zones | ACL changes on a compromised boundary device | T1599 Network Boundary Bridging |
Reachability: out-of-band networks and access control lists
The CISA joint guidance asks for an out-of-band (OOB) management network that is physically separate from the data network, with device management allowed only from it, no management from the internet, and administration done from dedicated workstations in a management zone. It also asks operators to prevent lateral management connections between devices and to disable outbound connections from devices, then alert on changes that re-enable them.
- Dedicated management interface: Bind management services to a management port, a loopback in a management VRF, or an OOB interface. On Cisco IOS, Management Plane Protection restricts the interfaces that accept management traffic, for example
control-plane hostthenmanagement-interface GigabitEthernet0/1 allow ssh https. Cisco documents MPP as IPv4 only, so use Control Plane Policing where IPv6 is in use. - VTY and service ACLs: Apply an ACL to VTY lines (
access-class) and to SNMP, HTTPS and NTP services that permits only named jump hosts, monitoring servers and AAA servers. Log denied attempts. - Infrastructure ACLs: Cisco's IOS hardening guide describes infrastructure ACLs that permit only required traffic to infrastructure addresses, such as SSH on TCP 22 from management stations and SNMP on UDP 161 from monitoring servers, and deny everything else addressed to the devices while allowing transit traffic.
- No outbound sessions:
transport output noneon VTY lines stops a device being used as a hop to the next device, which is how attackers chain through a network. - Firewall and appliance equivalents: Fortinet's FortiGate hardening guidance advises changing default administrative ports and the default administrator username, avoiding administrative access on WAN interfaces, restricting administrators to trusted hosts and using local-in policies, with logging, to control which hosts reach admin ports.
For US federal civilian agencies, CISA Binding Operational Directive 23-02 (13 June 2023) requires that, within 14 days of discovery, a networked management interface on a router, switch, firewall, VPN concentrator, proxy, load balancer or out-of-band server controller such as iLO or iDRAC is either removed from the internet or placed behind a Zero Trust policy enforcement point separate from the interface. The protocols in scope include HTTP, HTTPS, FTP, SNMP, Telnet, TFTP, RDP, rlogin, RSH, SSH, SMB, VNC and X11. Design detail for the separate network is in the Out-of-Band Management Network Design Guide, and admin endpoint design in the Privileged Access Workstations Guide.
Protocol baseline: what to disable, what to keep
The CISA joint guidance names the protocols to remove and the cryptographic floor for the ones that remain. The table combines that guidance with the Cisco IOS hardening guide and the IETF TACACS+ documents.
| Function | Remove | Use instead | Example setting |
|---|---|---|---|
| Interactive CLI | Telnet (TCP 23), SSH version 1 | SSH version 2 only; RSA key of at least 3072 bits, Diffie-Hellman group 16 (CISA) | ip ssh version 2; transport input ssh; transport output none |
| Web management | HTTP (TCP 80) | HTTPS with TLS 1.3 and a PKI certificate, or no web server at all | no ip http server; no ip http secure-server if the GUI is not needed |
| File and config transfer | FTP, TFTP, configs as email attachments | SCP or SFTP over SSH | ip scp server enable |
| Monitoring | SNMP v1 and v2c community strings | SNMPv3 with authentication and privacy (authPriv), ACL restricted | snmp-server group NAME v3 priv with an access list |
| Device AAA | TACACS+ relying on body obfuscation over shared networks | TACACS+ over TLS 1.3 (RFC 9887, TCP 300) where supported, otherwise TACACS+ only across the protected management network | Key stored as Type 6 where supported (CISA) |
| Unauthenticated management | Cisco Smart Install, guest shell where not needed, auto-install of config or image | Controlled provisioning from the configuration management system | no vstack; guestshell disable; FortiOS set auto-install-config disable |
| Discovery and legacy services | CDP and LLDP on untrusted interfaces, IP source routing, TCP and UDP small servers | Enable discovery only on interfaces that need it | no cdp run or per interface; no ip source-route |
| Time and routing integrity | Unauthenticated NTP, OSPF, BGP, HSRP | Authenticated sessions where supported | ntp authenticate with ntp trusted-key |
For VPN and TLS cryptography the CISA guidance names Diffie-Hellman groups 15, 16 or 20, AES-256 and SHA-384 or SHA-512, and asks operators to disable unused VPN features and weak algorithms. Supporting detail is in the SNMP Security Guide, the NTP Security Guide and the BGP and Routing Security Guide.
Administrator authentication and AAA design
RFC 8907, the Informational RFC that documents TACACS+ (September 2020), states that the protocol's MD5-based body obfuscation provides no meaningful integrity, privacy or replay protection, and that TACACS+ must be deployed over networks that ensure privacy and integrity. RFC 9887 (December 2025, Standards Track) updates it with TACACS+ over TLS 1.3 on TCP port 300, forbids obfuscation inside TLS, and makes certificate-based mutual authentication mandatory to implement. Check platform support before planning on it.
Central AAA
- Separate identity store: CISA advises that the device AAA server supports MFA and is not linked to the primary corporate identity store. If an attacker takes the corporate directory, they should not inherit every router.
- Phishing-resistant MFA: The guidance asks for hardware-based PKI or FIDO methods for router administration. In practice this is usually enforced at the jump host or privileged access gateway that brokers SSH to devices.
- Per-command authorisation and accounting: With TACACS+, authorise and account every privileged command, for example
aaa authorization commands 15 default group tacacs+ noneandaaa accounting commands 15 default start-stop group tacacs+. Cisco notes command accounting is not supported with RADIUS, which is one reason TACACS+ remains common for device administration. See the RADIUS Security Guide. - Roles and least privilege: Map AAA groups to device privilege levels or roles: read-only for NOC staff, change rights for engineers, full rights only through just-in-time elevation.
Local and emergency accounts
- Emergency only: Keep one local account per device for AAA outage, store its password in a vault, change it after each use and confirm every use was authorised, as the CISA guidance asks. See the Break-Glass Account Guide.
- Strong hashing: Use
enable secretandusername NAME algorithm-type scrypt secret(Type 9) or Type 8. CISA advises Type 8 where possible and avoiding Type 5 and Type 7. Cisco's hardening guide describesservice password-encryptionas a Vigenere cipher that only stops casual viewing, so any Type 7 value in a configuration should be treated as clear text. - Lockout and timeouts:
aaa local authentication attempts max-faillimits guessing against local accounts (Cisco notes privilege 15 users are not locked out),exec-timeouton console and VTY lines closes idle sessions, andip ssh time-out 60withip ssh authentication-retries 3limits slow attempts. - Defaults: Never leave or reset to a default password, and change default administrator names where the platform allows it.
Where identity fits: device secrets as non-human credentials
A network estate holds thousands of credentials no human logs in with: SNMP credentials, AAA shared keys, NTP and routing keys, SSH keys, and the tokens and service accounts of automation tools. They are long-lived, often identical across hundreds of devices and present in configuration backups. Treat them like any other machine identity.
| Credential | Typical weakness | Control |
|---|---|---|
| SNMP community string | Same string everywhere, sent in clear text, write access enabled | Replace with SNMPv3 authPriv users; remove RW; restrict with ACL and view |
| TACACS+ or RADIUS shared key | One key across the estate, recoverable from traffic or config | Per site or per device keys, Type 6 storage, TLS transport where supported, rotation after any suspected exposure |
| Local admin password | Type 7 or Type 5 storage, never rotated | Type 8 or 9, vaulted, rotated after use |
| Automation service account and API token | Full privilege, no expiry, stored in scripts | Scoped role, secrets manager, short lifetime, source address restriction |
| SSH keys for admins and tools | Unmanaged authorised keys added to devices | Central key or certificate management; alert on new keys |
See the SSH Key Management Guide and the Vulnerability Management for Network Devices Guide for credential handling in scanners and automation tools.
Configuration integrity, images and change control
- Off-device source of truth: The CISA guidance says to store configurations centrally, push them to devices and not treat the device as the source of truth, then compare and override regularly. A drift report is both a hardening control and a detection.
- Protected backups: Configuration files contain hashes, keys and topology. Transfer them only over SCP, SFTP or HTTPS, encrypt them at rest and restrict who can read the backup repository.
- Rollback ready: Cisco's
archivefeature keeps numbered configuration copies andconfigure replacerestores a known version.configuration mode exclusive autostops two administrators changing configuration at once. - Image verification: Compare software image hashes with vendor-published values from an authenticated source before installation and periodically afterwards, and use the vendor's runtime integrity checks where available. Watch end-of-life notices and maintain a device and firmware inventory.
- Change review: Require a ticket and peer review for changes to AAA, ACLs, SNMP, logging targets and local accounts, because these are exactly the changes an attacker makes. See the Configuration Drift and Baseline Enforcement Guide.
Logging, detection and measurement
The CISA guidance asks for device logging sent to a central system over an encrypted transport such as IPsec or TLS, off-site copies that cannot be modified or deleted, and alerts on unauthorised configuration changes, unusual route updates, weak protocols being enabled, and changes to users and ACLs. On Cisco IOS the building blocks are logging host, logging source-interface Loopback0, logging trap at an informational level, service timestamps log datetime msec show-timezone and authenticated NTP so timestamps can be trusted. Detection rules for these sources are covered in the Network Device Syslog and Configuration Change Detection Guide.
High-value alerts
- AAA tampering: Any change to TACACS+ or RADIUS server addresses or keys, or AAA method lists that add a local fallback.
- New persistence: Local account creation, privilege changes, new SSH authorised keys, SSH enabled on a non-standard port.
- Protocol regression: Telnet, HTTP, SNMP v2c or Smart Install appearing in a running configuration.
- Unusual access: Successful logins from outside the jump host range, logins outside change windows, use of the emergency account.
- Configuration exfiltration: TFTP or FTP flows from devices, and SNMP queries for configuration objects from unexpected sources.
Metrics
- Exposure: Percentage of devices whose management services accept connections only from the management network, verified by external and internal scans.
- Protocol hygiene: Count of devices with Telnet, HTTP, SNMP v1/v2c or Type 7 passwords in the running configuration (target zero).
- Central AAA coverage: Percentage of administrative logins authenticated through central AAA with MFA rather than local accounts.
- Drift: Number of unapproved configuration differences found per week and median time to resolve.
- Secret age: Age of the oldest TACACS+ key, SNMPv3 credential and emergency password.
- Logging completeness: Percentage of devices sending logs in the last 24 hours.
Implementation roadmap
- Inventory and scan (weeks 1 to 2): Build a list of every device, OS version and management service. Scan from the internet and from user VLANs for SSH, HTTPS, SNMP and Telnet responses. Anything reachable from the internet is fixed first.
- Restrict reachability (weeks 2 to 4): Apply VTY and service ACLs permitting only jump hosts and monitoring servers, then MPP or equivalent. Test on a pilot set of devices with console access available, and keep the previous configuration archived so
configure replacecan roll back a lockout. - Remove plaintext protocols (weeks 3 to 6): Move monitoring to SNMPv3 before removing v2c, move backups to SCP before disabling TFTP, and confirm tools work after each change. Disable Smart Install, HTTP and unused services.
- Rebuild AAA (weeks 4 to 10): Stand up or separate the device AAA service from the corporate directory, put MFA in front of administration at the jump host, enable command authorisation and accounting, and set per-site keys. Keep one vaulted emergency account.
- Rotate secrets (weeks 6 to 12): Rotate every SNMP credential, AAA key, local password and automation token, convert to Type 8 or 9 hashing, and record owners and rotation dates.
- Centralise configuration and logging (ongoing): Make the configuration repository authoritative, run nightly drift comparisons, ship logs over an encrypted transport to immutable storage, and deploy the alerts above.
Decision guidance
| Decision | Option A | Option B | How to choose |
|---|---|---|---|
| OOB design | Physically separate management network with its own switches and console servers | Logical management VRF or VLAN over the production network | Physical OOB for service providers, data centres and critical sites; logical separation with strict ACLs is acceptable for small branches if management traffic never crosses user segments |
| AAA protocol | TACACS+ | RADIUS | TACACS+ when you need per-command authorisation and accounting; RADIUS when the platform or team already relies on it, with command logging gathered from syslog instead |
| MFA placement | At a privileged access gateway or jump host | At the AAA server for every device login | Gateway MFA is easier to deploy and also gives session recording; AAA MFA covers direct logins if the gateway is bypassed. Large estates often use both |
| Small organisation | With a handful of devices, the minimum is: no internet-exposed management, SSHv2 and HTTPS only from one admin host, SNMPv3 or no SNMP, strong local password hashing, a vaulted emergency account and logs shipped to the SIEM or a managed service. | ||
Common failure patterns
- Management on the user network: Devices accept SSH from any internal address, so one compromised laptop can reach every router.
- AAA tied to the corporate directory: A directory compromise or a phished engineer gives direct device access, with no separate barrier.
- Configurations as secret stores: Backups sent over TFTP or stored in a shared folder hold Type 7 passwords and SNMP strings that can be reversed offline, which is what public reporting on the Salt Typhoon telecom intrusions describes.
- One shared key everywhere: A single TACACS+ key or community string across the estate means one capture exposes all devices, and rotation becomes so painful it never happens.
- Logs only on the box: Without off-device, immutable logs, an attacker who clears local logs removes the evidence, and investigators cannot reconstruct changes.
- Patching without checking: Upgrading a device after a known exploit without checking for added accounts, keys or changed AAA leaves the attacker's persistence in place. See the Edge Device Compromise Investigation Playbook.
Framework mapping
| Control area | NIST SP 800-53 Rev. 5 | NIST CSF 2.0 | CIS Controls v8.1 | ISO/IEC 27001:2022 Annex A |
|---|---|---|---|---|
| Management reachability and segmentation | SC-7, AC-17, MA-4 | PR.IR-01 | 12.3, 12.8 | 8.20 Networks security |
| Administrator authentication and roles | IA-2, AC-2, AC-6 | PR.AA-01, PR.AA-05 | 5.4, 6.5 | 8.2 Privileged access rights; 8.5 Secure authentication |
| Protocol and service minimisation | CM-7, SC-8, SC-13 | PR.PS-01 | 4.6, 4.8 | 8.9 Configuration management |
| Device secrets | IA-5 | PR.AA-01 | 12.3 | 5.17 Authentication information |
| Configuration and image integrity | CM-3, CM-6, SI-7 | PR.PS-01 | 4.1, 4.2 | 8.9 Configuration management; 8.32 Change management |
| Logging and monitoring | AU-12, AU-9 | DE.CM-01 | 8.2, 8.9 | 8.15 Logging; 8.16 Monitoring activities |
Related CIS guidance is in the CIS Control 12 Network Infrastructure Management Guide.
Practitioner checklist
- Scan from the internet and from user networks for device management services, and close every exposure found.
- Restrict SSH, HTTPS and SNMP on every device to named jump hosts and monitoring servers with ACLs, and disable outbound VTY sessions.
- Remove Telnet, HTTP, FTP, TFTP, SSHv1, SNMP v1/v2c and Smart Install, and keep them out with drift alerts.
- Separate device AAA from the corporate directory and require phishing-resistant MFA for administration.
- Enable per-command authorisation and accounting, and send AAA logs to the central log platform.
- Convert local passwords to Type 8 or Type 9, vault one emergency account per device and rotate it after each use.
- Inventory and rotate every SNMP credential, AAA key and automation token, and record an owner for each.
- Make the configuration repository authoritative, run nightly drift checks and verify image hashes against vendor values.
- Alert on AAA server changes, new local accounts or SSH keys, protocol regressions and logging silence.
Standards and references
- CISA, NSA, FBI and partners: Enhanced Visibility and Hardening Guidance for Communications Infrastructure
- CISA: Binding Operational Directive 23-02, Mitigating the Risk from Internet-Exposed Management Interfaces
- IETF: RFC 9887, TACACS+ over TLS 1.3
- IETF: RFC 8907, The Terminal Access Controller Access-Control System Plus (TACACS+) Protocol
- Cisco: Guide to Harden Cisco IOS Devices
- Fortinet: FortiGate Best Practices, Hardening
- MITRE ATT&CK: Network Device Authentication (T1556.004)
- NIST: SP 800-53 Rev. 5
Related NHI Mgmt Group resources: Edge Device Security Guide · Out-of-Band Management Network Design Guide · Network Device Syslog and Configuration Change Detection Guide · SNMP Security Guide · Salt Typhoon Threat Actor Profile