Accurate, trustworthy time is a security dependency that most organisations never secure: plain NTP is unauthenticated UDP, so anyone on the path can shift a clock, and a shifted clock quietly breaks Kerberos, certificate validation, token expiry and every log timeline an investigation relies on. This guide covers the threats to time synchronisation, how Network Time Security (NTS, RFC 8915) closes the authentication gap, how to design a resilient time hierarchy, and how to harden ntpd, NTPsec, chrony and Windows Time. It also covers NTP as an amplification source, monitoring and the framework controls that require synchronised clocks. It is written for network, infrastructure and security engineers who own time services or depend on them for authentication and detection.
Key takeaways
- Time is an authentication input. Kerberos tickets, TLS certificate validity, OAuth and SAML token lifetimes, TOTP codes and log correlation all assume the clock is right, so an attacker who controls time controls those checks.
- NTS is the modern answer for client-server NTP. RFC 8915 (September 2020, Standards Track) uses TLS 1.3 on TCP port 4460 for key establishment, then authenticates every NTP packet on UDP 123 with AEAD_AES_SIV_CMAC_256.
- Authentication does not stop delay attacks. RFC 8915 itself says an on-path attacker can add asymmetric delay and cryptography offers no practical mitigation, so diversity of sources is still required.
- Four or more diverse sources is the baseline. RFC 8633 (BCP 223) recommends at least four independent sources for operators who care about accurate time, which lets the selection algorithm outvote a bad one.
- Public-facing NTP must not answer control queries. The monlist command (CVE-2013-5211) turned unpatched ntpd into a large DDoS amplifier; block mode 6 and 7 queries from outside and use noquery restrictions.
Why time integrity is a security control
NTPv4 (RFC 5905) runs over UDP port 123 and in its default form carries no authentication. A client accepts whatever the selected server says, within the limits its discipline algorithm allows. That matters because many controls compare a timestamp to the local clock:
- Kerberos: Active Directory rejects authenticators outside the maximum tolerance for computer clock synchronisation, 5 minutes by default. A domain controller pushed out of step causes authentication failures across the estate. See the Kerberos Security Guide.
- Certificates and revocation: moving a client clock backwards can make an expired or revoked certificate look valid again, and moving it forward can make every valid certificate look expired. With TLS certificate lifetimes shrinking, clock error turns into outages faster. See the Certificate Expiry Outages and 47-Day TLS Certificates Guide.
- Tokens and one-time codes: JWT exp and nbf claims, SAML NotOnOrAfter conditions and TOTP windows all depend on the verifier's clock. A rolled-back clock can extend the life of a stolen token. See the Token Lifetime Policy Design Guide.
- Logs and forensics: correlating events across firewalls, identity providers, endpoints and cloud audit trails depends on consistent timestamps. NIST SP 800-53 AU-8 (Time Stamps) and SC-45 (System Time Synchronization) exist for this reason. See the Log Management and Retention Guide.
- Scheduled and replay-sensitive operations: backup retention, key rotation jobs, certificate renewal, licence checks and replay windows in many protocols assume monotonic, correct time.
Threats to time synchronisation
| Threat | How it works | Effect | Primary control |
|---|---|---|---|
| Spoofed or modified NTP replies | On-path or adjacent attacker forges server responses on UDP 123 | Client clock shifted, often gradually to avoid step limits | NTS or symmetric-key authentication; multiple sources |
| Delay attack | Attacker holds packets in one direction to create asymmetric delay | Offset error bounded by half the round-trip delay; not stopped by crypto | Diverse network paths and sources; offset alarms |
| Rogue or compromised upstream | Client uses an unvetted public server, or a source is hijacked via DNS or BGP | Wrong time served with valid-looking responses | Vetted sources, NTS with certificate checks, minsources quorum |
| Kiss-o'-Death abuse | Spoofed KoD packets tell a client to stop polling a server | Client loses good sources and falls back to worse ones | Accept KoD only with a valid origin timestamp (RFC 8633); NTS-protected sessions |
| NTS stripping | Attacker blocks TCP 4460 so the client gives up on NTS | Silent fallback to unauthenticated NTP | Never configure automatic fallback; alert on NTS-KE failure |
| Reflection amplification | Spoofed mode 7 monlist or mode 6 queries to open servers | Your server floods a third party (ATT&CK T1498.002) | noquery, upgraded daemons, egress and ingress filtering |
| GNSS spoofing or jamming | Radio interference with the GPS or other GNSS receiver feeding a stratum 1 server | Whole hierarchy inherits wrong time or loses its reference | Multiple reference types, holdover oscillators, cross-checks against network sources |
| Local clock tampering | Administrator-level attacker changes system time or file timestamps (T1070.006 Timestomp) | Misleading forensic timelines | Restrict time-change privilege, central logging with receive-time stamps |
Attackers also read the clock: ATT&CK T1124 System Time Discovery covers malware checking local time and time zone, for example to delay execution. That is a detection signal, not a reason to hide time.
How Network Time Security works
NTS, specified in RFC 8915, has two parts.
- NTS Key Establishment (NTS-KE): the client opens a TLS connection to the server on TCP port 4460, using ALPN identifier "ntske/1". Implementations must not negotiate anything below TLS 1.3. The client validates the server's X.509 certificate against the hostname, so NTS servers are addressed by name, not IP address. Both sides derive client-to-server and server-to-client keys using TLS key export with the label "EXPORTER-network-time-security", and the server hands back a set of opaque cookies.
- NTS extension fields for NTPv4: each NTP request carries a Unique Identifier, one cookie and an NTS Authenticator and Encrypted Extension Fields field. The server rebuilds the keys from the cookie, so it keeps no per-client state, and returns a fresh cookie in every response. AEAD_AES_SIV_CMAC_256 (numeric identifier 15) is mandatory to implement.
Design properties that matter operationally:
- What it protects: server identity, packet integrity and authenticity, request and response matching, replay detection for clients, and confidentiality of extension fields.
- What it does not protect: the NTP header is still sent in the clear; delay attacks remain possible; unlinkability applies only to time packets, not key establishment; and an attacker can still send unauthenticated spoofed KoD packets to push load onto NTS-KE servers.
- Non-amplification: responses are no larger than requests, so NTS servers are poor reflectors.
- No silent downgrade: RFC 8915 says clients should not fall back to plain NTP without user action. Treat fallback as an alarm.
NTS compared with older NTP authentication
| Option | Key management | Strengths | Limitations | Use when |
|---|---|---|---|---|
| Unauthenticated NTP | None | Universal support | Trivially spoofed on path | Only as one of several diverse sources, never alone for critical systems |
| Symmetric keys (ntp.keys, trustedkey) | Pre-shared key per association, distributed manually | Simple, works on many network devices | Key distribution and rotation by hand; does not scale to large client populations | Router and switch fleets talking to your own servers; symmetric peers |
| Autokey (RFC 5906) | Public-key scheme for NTPv4 | Historical | Known cryptographic weaknesses; not supported in NTPsec or chrony | Do not deploy |
| Windows domain hierarchy (NT5DS) | Kerberos session key of the computer account | Automatic inside an AD forest | Only inside the domain; the forest root PDC emulator still needs a trustworthy upstream | All domain-joined Windows systems |
| NTS (RFC 8915) | Automatic via TLS 1.3 and web PKI or a private CA | Scales, rotates keys, stateless servers, certificate-based identity | Needs a client that supports it (chrony 4.0 or later, NTPsec 1.2.0 or later); TCP 4460 must be open; no delay protection | Linux and appliance clients reaching internal or public NTS servers |
Designing a resilient time hierarchy
Most enterprises should run a small internal time service and point everything at it, rather than letting thousands of hosts reach public servers.
- Choose references: combine at least two reference types where time is critical, for example GNSS-disciplined stratum 1 appliances with good holdover oscillators plus NTS-authenticated public or national time services. Only synchronise to public servers you have permission to use: RFC 8633 notes the NTP Pool is volunteer-run and that vendors should obtain their own pool subdomain.
- Build a stratum 2 tier: three or four internal servers per region, each using at least four upstream sources (RFC 8633 Section 3.2), spread across providers and network paths so one hijacked route cannot move all of them.
- Serve clients internally: clients use three or four internal servers. Use NTS where clients support it; use symmetric keys for network devices that do not.
- Integrate Active Directory: the PDC emulator in the forest root domain is the best time source for the forest unless another reliable source is configured; other domain controllers and members follow the domain hierarchy and authenticate NTP with the computer's Kerberos session key. Point the forest root PDC emulator at your internal stratum 2 servers, not at a single public name, and remember that the role moves when the PDC emulator is transferred.
- Handle virtualisation deliberately: decide whether guests take time from the hypervisor integration service or from NTP, not both, and make sure hypervisor hosts themselves are synchronised to the same hierarchy.
- Agree a leap second policy: RFC 8633 says public-facing servers should not leap smear and clients must not mix smeared and non-smeared sources. Pick one approach estate-wide.
Small versus large organisations
- Small organisation: two internal chrony or NTPsec servers using four or more NTS-capable public sources, domain controllers syncing from them, network devices using symmetric keys. No GNSS hardware needed unless regulation demands traceability.
- Large or regulated organisation: GNSS-disciplined appliances in at least two sites, internal NTS service with a private CA, per-region stratum 2 servers, offset monitoring in the SOC and documented traceability to UTC where financial or telecoms rules require it.
- OT and telecoms: time often comes from GNSS or PTP rather than NTP, so the GNSS antenna and receiver become part of the attack surface. See the Telecom Network Security Guide.
Hardening NTP servers and clients
| Platform | Setting | Purpose |
|---|---|---|
| chrony | server time.example.net iburst nts | Enable NTS for a source (off unless specified) |
| chrony | ntsservercert and ntsserverkey | Act as an NTS server; the NTS-KE port (ntsport, default 4460) opens only when both are set |
| chrony | minsources 2 or higher | Require a quorum before updating the clock (default 1) |
| chrony | maxdistance, maxchange, makestep | Reject distant sources (default 3 s root distance), alarm or exit on large offsets, allow stepping only in the first updates after boot |
| chrony | allow and deny | Serve only named subnets; by default no clients are allowed |
| chrony | cmdport 0 or default loopback binding | Keep the monitoring port (UDP 323) off the network |
| chrony | ratelimit, ntsratelimit | Limit per-client NTP and NTS-KE rates on servers |
| NTPsec | server name nts; nts enable, nts cert, nts key, nts cookie | NTS client and server configuration; cookie keys default to /var/lib/ntp/nts-keys |
| ntpd and NTPsec | restrict default -4 nomodify notrap nopeer noquery and the -6 equivalent | Block configuration changes, traps, peering and mode 6 or 7 queries from everyone else |
| ntpd | Version 4.2.7 or later | monlist removed; older versions answer it by default |
| Windows | w32tm /config /manualpeerlist:"..." /syncfromflags:manual /update on the forest root PDC emulator only | Set the forest's upstream sources; other DCs keep the domain hierarchy |
| Windows | MaxPosPhaseCorrection and MaxNegPhaseCorrection under HKLM\SYSTEM\CurrentControlSet\Services\W32Time\Config | Cap how far W32Time will move the clock in one correction |
| Network devices | NTP authentication keys, source interface, access group limiting peers and queries | Stop devices from serving or accepting time from arbitrary hosts |
- Least functionality: do not run a time server on hosts that only need to be clients. Disable symmetric passive associations except with trusted, authenticated peers (RFC 8633 Section 5.6).
- Kiss-o'-Death handling: clients should back off on KoD but accept it only with a valid origin timestamp, and should not blindly adopt a server-suggested poll interval.
- Privilege to change time: on Windows, restrict the "Change the system time" user right (SeSystemtimePrivilege) to LOCAL SERVICE and administrators; on Linux, limit CAP_SYS_TIME to the time daemon.
- Firewalling: allow UDP 123 and TCP 4460 only between clients and their internal servers, and between internal servers and named upstreams. See the Firewall Management Guide.
NTP as a DDoS reflector
CISA's alert on CVE-2013-5211 describes attackers sending monlist requests with a spoofed source address; monlist returns up to the last 600 addresses that contacted the server, so the response is far larger than the request and lands on the victim. All ntpd versions before 4.2.7 were vulnerable by default. The fix is to upgrade, or to add noquery to the restrict default lines for both IPv4 and IPv6. The same logic applies to mode 6 (ntpq) queries: RFC 8633 says public-facing servers should block control queries from outside the organisation and require authentication (controlkey or requestkey) for anything beyond basic monitoring. Network-wide, source address validation at your edge stops your own hosts from being used to spoof. Map this to ATT&CK T1498.002 Reflection Amplification and mitigation M1037 Filter Network Traffic.
Implementation roadmap
- Inventory: find every configured time source (chrony.conf, ntp.conf, W32Time registry, network device configs, appliances, cloud images) and every host answering on UDP 123. Flag any device using a single source or a hard-coded public IP.
- Stand up the internal tier: deploy two to four stratum 2 servers with at least four diverse upstreams, NTS where upstreams support it, restrict and allow lists, and monitoring.
- Pilot NTS: enable
ntson a pilot group of Linux servers alongside existing sources; confirm NTS-KE succeeds (chrony:chronyc -N authdata; NTPsec:ntpq -c ntsand the cookie count column inntpq -p, which should be 8). - Move AD: repoint the forest root PDC emulator to the internal tier, verify with
w32tm /query /statusandw32tm /query /source, then check a sample of DCs and members. - Network devices: configure symmetric-key authentication to internal servers in waves, starting with core routers and firewalls.
- Close the old paths: block outbound UDP 123 from clients to the internet, block inbound control queries, and remove time server roles from hosts that do not need them.
- Rollback: keep the previous source lines commented in configuration management; if NTS-KE fails broadly, revert the pilot group by configuration push rather than enabling automatic fallback.
Detection and measurement
- Offset and stratum monitoring: collect offset, jitter, stratum and reachability from every internal server (
chronyc tracking,chronyc sources,ntpq -p,w32tm /stripchart) and alert on sudden steps. - Source disagreement: alert when one source is rejected as a falseticker or when the selected source changes repeatedly. RFC 9523 (Khronos, February 2024, Informational) describes a watchdog that samples many servers and treats an offset above about 30 ms as a likely time-shifting attack; the idea is useful even without deploying it.
- Clock change events: collect Windows Security event 4616 (system time was changed) and Linux audit rules on clock_settime, settimeofday and adjtimex, and correlate with the time daemon's own steps.
- NTS health: alert on NTS-KE handshake failures, certificate errors and cookie exhaustion; these are the signal for stripping attempts.
- Exposure: scan your external ranges for UDP 123 responders and for mode 6 or 7 replies.
Metrics that show the programme is working:
- Percentage of hosts synchronised to approved internal sources (target 100 percent).
- Percentage of capable clients using NTS or symmetric-key authentication.
- 95th percentile absolute offset across servers versus your tolerance (for example under 10 ms for general IT).
- Count of internet-reachable hosts answering NTP control queries (target zero).
- Number of internal time servers with fewer than four healthy upstreams.
- Mean time to detect an injected offset in a test.
Framework mapping
| Framework | Reference | Relevance |
|---|---|---|
| NIST SP 800-53 Rev. 5 | SC-45, SC-45(1) Synchronization with Authoritative Time Source | Synchronise clocks to an authoritative source |
| NIST SP 800-53 Rev. 5 | AU-8 Time Stamps | Audit records use internal clocks mapped to UTC |
| NIST SP 800-53 Rev. 5 | CM-7 Least Functionality; SC-5 Denial-of-service Protection | No unnecessary time server roles; no reflection |
| NIST CSF 2.0 | PR.PS-04; DE.CM-01 | Log records available for monitoring; networks monitored |
| CIS Controls v8.1 | Safeguard 8.4 Standardize Time Synchronization | At least two synchronised time sources across assets |
| ISO/IEC 27001:2022 Annex A | A.8.17 Clock synchronization | Clocks synchronised to approved sources |
| MITRE ATT&CK v20 | T1498.002; T1124; T1070.006 | Reflection amplification; time discovery; timestamp tampering |
Common failure patterns
- Single upstream: the forest root PDC emulator points at one public hostname, so one bad answer or one outage moves or freezes the whole domain.
- PDC emulator moved, time config not: the role is transferred during a migration and the new holder still uses the domain hierarchy, leaving the forest with no external reference.
- Silent NTS fallback: NTS is configured alongside plain servers, TCP 4460 is blocked by a new firewall rule, and nobody notices that clients are now unauthenticated.
- Exposed legacy daemons: appliances and old Unix hosts still answer monlist or mode 6 queries from the internet and become reflectors.
- Mixed smearing: some sources leap smear and others do not, and clients see a disagreement of up to a second around a leap event.
- Logs without a time standard: devices log in local time with no zone, or with drifting clocks, and incident timelines cannot be reconciled. See the Digital Forensics (DFIR) Guide.
Practitioner checklist
- Inventory every time source and every host serving NTP, including network devices and appliances.
- Run an internal stratum 2 tier with at least four diverse, permitted upstream sources per server.
- Enable NTS (RFC 8915) for every client and upstream that supports it, and alert rather than fall back on NTS-KE failure.
- Use symmetric-key authentication for network devices that cannot do NTS; never deploy Autokey.
- Point the forest root PDC emulator at the internal tier and re-check after any role transfer.
- Apply
noquery,nomodify,notrapandnopeerrestrictions, and block NTP control queries at the perimeter. - Set quorum and sanity limits (chrony
minsources,maxchange; W32Time phase correction caps). - Restrict who can change system time and collect clock change events centrally.
- Monitor offset, source disagreement and NTS health, and report the metrics above monthly.
Standards and references
- IETF: RFC 8915 Network Time Security for the Network Time Protocol
- IETF: RFC 8633 Network Time Protocol Best Current Practices (BCP 223)
- IETF: RFC 9523 A Secure Selection and Filtering Mechanism for NTP with Khronos
- CISA: NTP Amplification Attacks Using CVE-2013-5211
- NIST: SP 800-53 Rev. 5
- chrony project: chrony.conf documentation
- NTPsec: NTS Quick Start Guide
- Microsoft Learn: How the Windows Time Service Works
Related NHI Mgmt Group resources: Network Security Guide · DNS Security Guide · DHCP Security Guide · Kerberos Security Guide · CIS Control 8 Audit Log Management Guide