Cisco Identity Services Engine decides who and what may join the network, and holds the RADIUS and TACACS+ secrets, Active Directory connections, certificates and API credentials to do it, so an attacker who controls ISE controls network access for the whole organisation. Apply Cisco's security updates promptly and track the Cisco Security Advisories page. This guide covers how identity and administration work in ISE, the admin roles and API groups, the non-human credentials it holds, logging and detection, and a hardening baseline. It is for network security engineers and identity teams running ISE, and goes deeper on the platform than our RADIUS Security Guide, which covers the protocol.
This is an independent guide from NHI Mgmt Group, based on Cisco's public documentation as of November 2026. It is not sponsored or reviewed by Cisco. Product features change, so confirm settings against Cisco's current documentation.
Key takeaways
- Patch ISE as an edge device. Apply Cisco's security updates promptly, and restrict management traffic to ISE with infrastructure ACLs.
- Keep the management plane unreachable. The admin portal runs on fixed ports 80 and 443 and APIs on 443, 9060 and 9070; only a management network and named jump hosts should reach them.
- Use the narrowest admin and API groups. ERS Operator is read-only; ERS Admin and Super Admin can change everything. Map AD groups into them and review membership.
- ISE is a vault of machine credentials. Per-device RADIUS and TACACS+ secrets, the AD join, pxGrid client certificates, backup encryption keys and repository passwords all need owners, rotation and storage outside spreadsheets.
- Send everything off-box. Forward the Administrative and Operational Audit category and authentication results to a SIEM through a secure syslog target, because an attacker with root on ISE can remove local evidence.
How identity and access work in Cisco ISE
ISE is a policy server. Network devices (switches, wireless controllers, VPN headends, firewalls) send RADIUS requests for network access and TACACS+ requests for device administration; ISE evaluates policy sets, authentication policies and authorisation policies, and returns a result such as a VLAN, a downloadable ACL or a TrustSec security group tag. A deployment is made of nodes with personas:
| Persona | Role | Security significance |
|---|---|---|
| Policy Administration Node (PAN) | Holds configuration; primary PAN replicates to all other nodes | Compromise of the primary PAN means compromise of every node's configuration |
| Monitoring and Troubleshooting (MnT) | Collects logs and session data, serves reports | Holds a history of every user and device authentication |
| Policy Service Node (PSN) | Answers RADIUS and TACACS+, runs portals (guest, BYOD, posture) | Exposed to the access network and, for guest portals, to untrusted users |
| pxGrid | Shares session and context data with other security tools | Gives subscribers user, device and IP mappings; can accept rapid threat containment actions |
Identity sources. ISE authenticates endpoints and users against internal identity stores, Active Directory join points, LDAP, SAML identity providers for portals, and certificate authentication profiles for EAP-TLS. Each AD join point gives ISE a computer account in the domain. Authorisation then uses AD group membership and endpoint profiling attributes, which means AD group changes and profiling data directly change network access.
Decision guidance. Prefer EAP-TLS with certificates from your PKI for managed devices; it removes reusable passwords from network access. Use PEAP with AD credentials only where certificates cannot be deployed, and MAC Authentication Bypass only for devices that cannot authenticate, combined with profiling and a restrictive authorisation result. Small deployments can run all personas on two nodes; large deployments should separate PAN and MnT from PSNs, keep PSNs in the network segments they serve, and keep PAN and MnT in the management zone. See the Network Access Control (NAC) Guide.
Admin roles and privileged access
ISE uses role-based administration with predefined admin groups. Cisco's documentation confirms these, among others:
| Admin group | Access | Assign to |
|---|---|---|
| Super Admin | All functions and all API services; only Super Admin can configure the admin account disable policy | Two to three named engineers and a vaulted break-glass account |
| Policy Admin | Authentication and authorisation policy configuration | Network access policy engineers |
| System Admin | System operations, including viewing authentication results | Platform operators |
| ERS Admin | Create, read, update and delete through the ERS API | Automation accounts that must write configuration |
| ERS Operator | Read-only (GET) ERS API access | Inventory and reporting integrations |
| MnT Admin | Monitoring REST APIs | SIEM or reporting integrations that query sessions |
- External admin identities: map AD groups to admin groups through the AD join point as identity source, so leavers lose access with their AD account. Keep one internal Super Admin account for break-glass, with a long random password in a vault and alerting on its use. ISE supports CAC authentication for administrators, which is a route to certificate-based admin login.
- Admin access settings: under Administrator Access Settings, configure the password policy, account disable policy, lock or suspend after failed logins, session timeout, maximum concurrent sessions, a login banner, and restrict administrative access to select IP addresses.
- CLI and console: the CLI admin account created at installation is separate from GUI admins. Restrict SSH to the management network and use the Diffie-Hellman key exchange settings Cisco documents for secure SSH.
- Device administration: if ISE also provides TACACS+ for network device admins, its policy decides who gets privilege 15 on every switch and router. Treat TACACS+ command sets and shell profiles as privileged configuration. See the Network Device Management Plane Hardening Guide.
Integrations, API tokens and service accounts
ISE holds more non-human credentials than most network systems. Inventory each one with an owner and a rotation plan.
| Credential | Where it lives | Risk | Control |
|---|---|---|---|
| RADIUS shared secrets | Each network device entry | One leaked secret shared across devices exposes every flow | Unique random secret per device; generate and store in a secrets manager; enable Message-Authenticator |
| TACACS+ shared secrets | Network device entries | Exposure of device admin sessions | Unique per device; restrict TCP 49 to management sources |
| AD join point | Computer account in AD | Directory reads for every user lookup; a target for lateral movement (MITRE ATT&CK T1078.002 Domain Accounts) | Join with a delegated account that can only create computer objects in a dedicated OU; do not store domain admin credentials |
| API admin users (ERS, OpenAPI) | Internal or AD identities in API groups | Configuration change at scale | Read-only ERS Operator by default; certificate-based API authentication where possible; disable unused APIs |
| pxGrid clients | Client certificates | Access to user, device and IP context; quarantine actions | Manual approval of new clients; per-client certificates; review subscriptions |
| Backup encryption key and repository credentials | Backup schedule and repository configuration | A backup contains the full configuration including secrets | SFTP repositories, strong key in a vault, access restricted to the backup server |
| System certificates (admin, EAP, portal, pxGrid) | Certificate store on each node | Expired EAP certificates break network access; a stolen admin key enables impersonation | Inventory expiry dates, renew ahead of time, schedule the application restart Cisco notes after admin certificate renewal |
API facts that matter. In ISE 3.4, the ERS API is disabled by default; Cisco documents ERS on HTTPS port 443 (9060 optional), OpenAPI on 443 and 9070 between nodes, TLS 1.1 and 1.2 for ERS but not TLS 1.0, a 60 second ERS idle session timeout and an optional CSRF check. Certificate-based API authentication works only with the Internal identity source, requires the admin username in the certificate CN and a rolling restart of each node. See the Machine Identity, PKI and Certificate Lifecycle Guide.
Logging and detection
- Remote logging: ISE ships with a UDP syslog target and a SecureSyslogCollector target. Send the Administrative and Operational Audit, AAA Audit, Accounting, Passive ID and Posture and Client Provisioning Audit categories to the SIEM over the secure target. Note that local logging of Passed Authentications is disabled by default.
- Admin activity: alert on admin logins from outside the jump host range, logins by the break-glass account, new admin users, changes to admin group membership and changes to admin access settings. The Server Administrator Logins report records lockouts; the Change Configuration Audit report records configuration changes, including who set event suppressions.
- Policy tampering: alert on changes to policy sets, authorisation results that grant broad access, new network device entries and changed shared secrets. An attacker who adds a rogue network device or a permissive rule gains network access without touching endpoints (MITRE ATT&CK T1556 Modify Authentication Process).
- Advisory indicators: where a Cisco advisory lists indicators, follow it; for example Cisco has advised reviewing the ise-kong
access.logon every node for suspicious usernames, and checking network and firewall logs for unexpected uploads to or downloads from external addresses, because exploitation can yield root and evidence may be removed. If compromise is suspected, Cisco recommends re-imaging nodes and restoring configuration from backup. - Outbound traffic from ISE: ISE nodes should only talk to known destinations (AD, DNS, NTP, PKI, SIEM, repositories, Cisco update services). Alert on anything else.
Metrics: days from Cisco critical ISE advisory to patched (target within your edge device SLA); number of Super Admin accounts; percentage of network devices with unique RADIUS secrets; percentage of endpoints authenticating with EAP-TLS rather than PEAP or MAB; certificates expiring in the next 60 days; percentage of nodes forwarding audit categories to the SIEM.
Hardening baseline
- Patch first: confirm every node is on a fixed release for the current critical advisories. Check the Cisco Security Advisories page for the fixed release for your train, and prefer the listed patch releases over hot patches. Release 3.0 is past End of Software Maintenance.
- Restrict the management plane: apply infrastructure ACLs so that only management subnets reach ports 443, 22, 9060 and 9070 on ISE nodes, and configure Restrict Admin Access to named IP ranges. Guest and BYOD portals on PSNs should be the only interfaces reachable from user networks.
- Admin identities: map AD groups to the narrowest admin groups, set password, lockout and session timeout policy, keep one vaulted internal break-glass account.
- APIs: leave ERS disabled unless needed, use ERS Operator for read-only integrations, prefer certificate-based API authentication, enable CSRF checking where clients support it.
- Secrets: rotate RADIUS and TACACS+ secrets to unique random values per device, enable Message-Authenticator handling, and move to RadSec where devices support it.
- Cryptography: enable FIPS mode where policy requires it, disable legacy TLS versions, and remove weak EAP methods such as EAP-MSCHAPv2 where certificates allow.
- Backups: schedule encrypted configuration and operational backups to an SFTP repository; Cisco recommends at least 100 GB repository space. Test restore, because Cisco's compromise guidance relies on re-image and restore.
- Logging: enable the secure syslog target and forward the audit categories before go-live.
Testing and rollback: patch the secondary PAN and a PSN first, confirm authentications succeed, then the primary PAN. For secret rotation and authentication method changes, use a pilot switch stack with monitor mode (open authentication with logging) before closed mode, and keep the previous secret on the device entry until both sides are updated.
| Framework | Reference | Fit |
|---|---|---|
| NIST SP 800-53 Rev. 5 | AC-2, AC-6, AC-17, AU-2, AU-6, IA-3, IA-5, SC-8, SI-2 | Admin accounts and least privilege, remote admin, logging and review, device authentication, secret management, transport protection, patching |
| NIST CSF 2.0 | PR.AA-03, PR.PS-02, DE.CM-09 | Hardware authenticated; software maintained; monitoring |
| CIS Controls v8.1 | Control 6, Control 7, Control 12, Control 13 | Access control, vulnerability management, network infrastructure management, network monitoring |
| MITRE ATT&CK | T1190 Exploit Public-Facing Application, T1556, T1078.002 | Exploitation of ISE APIs, policy tampering, domain account abuse |
Common misconfigurations
- Management interfaces reachable from user networks: PSN portals are meant to be reachable; the admin portal and APIs are not, yet flat routing often exposes them on every node.
- One RADIUS secret for every switch: copied from a template years ago and stored in configuration backups.
- Super Admin for automation: scripts and integrations given Super Admin or ERS Admin when ERS Operator or MnT Admin would do.
- Domain admin used for the AD join: instead of a delegated account limited to a computer object OU.
- Hot patch treated as the fix: confirm in the advisory that a hot patch covers every issue; otherwise move to the listed patch release.
- Logs only on the MnT node: after root compromise, local logs cannot be trusted.
Practitioner checklist
- Confirm every ISE node is on a fixed release for the current Cisco security advisories.
- Apply infrastructure ACLs and Restrict Admin Access so only management networks reach admin and API ports.
- Map AD groups to the narrowest admin groups and keep one vaulted break-glass Super Admin.
- Disable unused APIs; use ERS Operator for read-only integrations and certificate-based API authentication.
- Rotate RADIUS and TACACS+ secrets to unique random values per device, stored in a secrets manager.
- Join AD with a delegated account limited to a dedicated OU.
- Approve pxGrid clients manually and review their subscriptions.
- Run encrypted backups to SFTP and test a restore.
- Forward audit categories to the SIEM over secure syslog and alert on admin, policy and device changes.
- Track certificate expiry for admin, EAP, portal and pxGrid certificates.
Standards and references
- Cisco: Cisco ISE Unauthenticated Remote Code Execution Vulnerabilities (cisco-sa-ise-unauth-rce-ZAd2GnJ6)
- Cisco: Cisco ISE Authentication Bypass Vulnerability (cisco-sa-ISE-ABP-VNSW7Tn5)
- Cisco: ISE 3.4 Administrator Guide, Basic Setup
- Cisco: ISE 3.4 Administrator Guide, Maintain and Monitor
- Cisco: ISE 3.4 Administrator Guide, Segmentation
- Cisco: Security Advisories
- NIST: SP 800-53 Rev. 5 Security and Privacy Controls
Related NHI Mgmt Group resources: RADIUS Security Guide · Network Access Control (NAC) Guide · Edge Device Security Guide · Cisco IOS XE Hardening Guide