Most PKI estates run all three certificate enrolment protocols at once, ACME for web servers, SCEP for managed devices and EST for network gear, often because each team picked what its tooling offered rather than what the use case needed. The choice matters because the enrolment protocol decides who may ask the CA for a certificate, how the requester proves it holds the key, and whether anyone checks it is entitled to the name it asks for. This guide is a selection matrix: for each use case (public and internal web servers, Kubernetes, service-to-service workloads, managed laptops and phones, network equipment, IoT and legacy devices) it names the protocol to prefer, the conditions that make it safe, and when none of the three is the right answer. It is for PKI engineers, endpoint and MDM teams, platform engineers and NHI owners. Protocol internals are in the ACME and Automated Certificate Management Guide and the IoT and Device Certificate Provisioning (EST, SCEP) Guide; this page does not repeat them.
Key takeaways
- Pick by how the requester proves entitlement. ACME proves control of a name, EST relies on TLS client authentication or an existing identity, and SCEP relies on a shared challenge that something else must validate.
- ACME for anything with a DNS name. Public and internal web servers, load balancers and Kubernetes ingress fit ACME's domain validation model and its large client ecosystem.
- SCEP only behind an MDM that validates the request. SCEP is an Informational RFC; it is safe when the MDM issues a one-time, signed challenge that binds the expected subject, as Intune's NDES policy module does.
- EST for devices that already have an identity. Network equipment and IoT devices with a manufacturer or bootstrap certificate can re-enrol over mutually authenticated TLS, the Standards Track option for non-web devices.
- Attestation is moving devices towards ACME. Apple's ACME payload with Managed Device Attestation and the IETF device-attest-01 draft prove a genuine, specific device with a hardware-bound key, which no shared challenge can.
The short answer
Use ACME where the subject is a DNS name you control and the client can complete a challenge: public TLS, internal TLS through a private ACME CA, and Kubernetes through an ACME issuer. Use SCEP where an MDM or UEM platform is the enrolment authority and its client support leaves no stronger option, and only with dynamic, validated challenges. Use EST for routers, switches, wireless controllers, IoT gateways and industrial devices that can authenticate with an existing certificate, or with BRSKI for zero-touch onboarding. For short-lived service-to-service identity inside a platform, prefer a workload identity system such as SPIFFE over any of the three. For new Apple device deployments on supported versions, evaluate ACME with attestation before defaulting to SCEP.
The three protocols on the attributes that decide selection
| Attribute | ACME | EST | SCEP |
|---|---|---|---|
| Specification and status | RFC 8555, Standards Track, March 2019 | RFC 7030, Standards Track, October 2013 | RFC 8894, Informational, September 2020 |
| Transport | JSON over HTTPS, every request signed with the account key (JWS) | HTTPS (TLS 1.1 or later required; use 1.2 or 1.3) under /.well-known/est/ | HTTP, with CMS-protected PKCS #10 messages |
| What the requester proves | Control of each identifier through a challenge (http-01, dns-01, tls-alpn-01) | Identity through a TLS client certificate (recommended), HTTP Basic or Digest inside TLS, or TLS-SRP | Knowledge of a challengePassword for first enrolment; the existing certificate for renewal |
| Identifier types | dns in RFC 8555; IP and other types by extension; device identifiers in the device-attest draft | Any subject the CA policy allows | Any subject the CA policy allows |
| Who checks name entitlement | The protocol, through validation of each identifier | The EST server or RA policy, using the authenticated client identity | An external component (MDM, RA policy module); the protocol itself does not |
| Proof of possession binding | CSR signed by the certificate key, inside a request signed by the account key | tls-unique value in the CSR ties proof of possession to the TLS session | Renewal requests are signed with the old key, so possession of the new key is not proven |
| Server key generation | Not defined | Optional /serverkeygen; RFC recommends servers do not enable it by default | Not defined |
| Renewal model | New order; ARI (RFC 9773, June 2025) lets the CA suggest renewal windows | /simplereenroll authenticated with the current certificate | RenewalReq signed with the current certificate |
Status checked on the RFC Editor on 14 November 2026.
Selection matrix by use case
| Use case | Prefer | Acceptable | Conditions that make it safe |
|---|---|---|---|
| Public web servers, CDN and SaaS custom domains | ACME | Vendor-managed issuance built on ACME | Scoped DNS API credentials or delegated _acme-challenge zones; CAA with account binding; ARI-aware clients |
| Internal web servers, APIs and load balancers | ACME against a private CA | EST from a server with an existing certificate | Internal ACME server with external account binding per team; DNS-01 against internal zones |
| Kubernetes ingress and gateway certificates | ACME issuer in the cluster controller | Private CA issuer through the same controller | Account key Secret restricted to the controller; one ACME account per cluster or environment |
| Service-to-service workload identity | Workload identity system (SPIFFE X.509-SVIDs) or platform pod certificates | ACME or EST only where the workload has a stable DNS name | Attested issuance, short lifetimes, automatic rotation through a workload API |
| Windows laptops via MDM | SCEP through the MDM with validated challenges | PKCS profiles where the MDM generates and delivers keys | Signed, per-device challenge with expected subject and SAN; strong mapping SID tag for AD authentication |
| Apple devices via MDM | ACME payload with hardware-bound key and attestation (supported versions) | SCEP through the MDM | CA verifies the attestation; device serial or identifiers checked against MDM inventory |
| Android and other phones via MDM | SCEP through the MDM with validated challenges | PKCS profiles | As for Windows; short validity and revocation on retire or wipe |
| Routers, switches, firewalls, wireless controllers | EST with a manufacturer or bootstrap certificate | SCEP where the platform offers nothing else | Explicit trust anchors; RA authorises names per device; no static shared challenge |
| New network or IoT devices, zero-touch | BRSKI followed by EST | EST with a pre-provisioned bootstrap credential | Manufacturer identity checked against purchase records; owner registrar controls enrolment |
| Constrained IoT on CoAP | EST-coaps | Gateway enrolling on the device's behalf | DTLS with device certificate; gateway credentials governed as privileged NHIs |
| Printers, legacy appliances, older OT | Whatever the device supports, usually SCEP | Manual issuance with an inventory entry | Isolated enrolment endpoint, one-time challenge per device, compensating network controls |
The deep dives explain the mechanics of each option. See the SPIFFE and SPIRE Guide and the Kubernetes Pod Certificates and ClusterTrustBundles Guide for the workload rows.
ACME: threat model, architecture and failure modes
Threat model: an attacker who gains control of the evidence the CA trusts for a name (port 80 on the host, the DNS zone or its API credential) or who steals the ACME account key. RFC 8555's stated goal is automating the verification and issuance of domain-validated certificates, so the security of the issued certificate equals the security of name control plus account key custody.
Architecture in words: a client on or near the server holds an account key, places an order for one or more names, satisfies a challenge for each, submits a CSR and installs the result; renewal repeats the cycle on a schedule or when the CA's renewal information suggests. A private ACME server lets the same clients serve internal names, with external account binding tying each ACME account to a team and policy.
Failure modes:
- Zone-wide DNS tokens: a DNS-01 credential that can edit the whole production zone turns a compromised client into the ability to mint certificates for, and redirect, every name in it.
- Shared accounts: one ACME account across all clusters means one leaked Secret affects every environment. cert-manager describes the account key as your identity with your ACME provider, and notes that losing it prevents revocation of certificates issued under that account.
- Forcing ACME onto devices without names: phones and appliances without a stable DNS name cannot complete classic challenges; the result is usually a weak workaround. Use attestation-based ACME or another protocol.
EST: threat model, architecture and failure modes
Threat model: an attacker who reaches the EST endpoint and tries to obtain a certificate for a device identity they do not own, or who intercepts enrolment traffic. EST answers with TLS: the server must authenticate, the client should authenticate with a certificate, and the tls-unique value in the CSR binds proof of possession to that TLS session, which defeats relaying a CSR through a different connection.
Architecture in words: devices ship with a manufacturer certificate (IDevID) or receive a bootstrap credential; they fetch CA certificates from /cacerts, enrol with /simpleenroll using their existing identity, and renew with /simplereenroll. The EST server, often an RA in front of the issuing CA, maps the authenticated identity to the names the device may receive. /csrattrs tells clients what the CA expects in the request.
Failure modes:
- Basic authentication everywhere: HTTP Basic inside TLS with a shared username and password reduces EST to a shared secret, losing its main advantage.
- Server-side key generation by default: /serverkeygen means the private key leaves the server and is protected only by TLS unless CMS encryption is used; RFC 7030 recommends servers do not support it by default.
- Implicit trust anchors for client authentication: accepting any certificate chaining to a general-purpose root lets unrelated certificates authenticate. Configure explicit trust anchors for device identities.
SCEP: threat model, architecture and failure modes
Threat model: anyone who obtains a valid challengePassword, or reaches an endpoint configured without one, and requests a certificate for an identity of their choosing. RFC 8894 recommends a high-entropy, one-time challenge with a limited lifetime, and notes that unauthenticated GetCACaps responses allow downgrade attacks, which is why single DES and MD5 are forbidden.
Architecture in words: the MDM delivers a profile with the SCEP URL, the CA chain and a challenge; the device generates a key, builds a PKCS #10 request wrapped in CMS and sends it to the SCEP server, often NDES in front of an AD CS Enterprise CA. In Microsoft Intune the challenge is an encrypted, signed blob containing the expected subject and SAN; the Intune policy module on NDES compares the request with the blob and refuses mismatches, and direct browsing to the NDES URL returns 403.
Failure modes:
- Static challenges: a challenge reused across devices is a public enrolment key for anyone who reads a profile or a configuration guide.
- Name trust left to the client: with "Supply in the request" templates and no policy module, any challenge holder can ask for any subject, including privileged users.
- NDES on the issuing CA: Microsoft says not to run NDES on the CA server; an internet-reachable enrolment service on the CA puts the most sensitive key one bug away from exposure.
- Renewal without new-key proof: because renewal is signed with the old key, limit renewal to the same subject and keep validity short.
For the AD CS side of these risks, see the AD CS Security Guide.
How hardware attestation changes the choice
Shared challenges and even existing certificates prove only that the requester holds a secret. Attestation proves the request came from a genuine, specific device with a key that cannot leave its hardware.
- Apple Managed Device Attestation: supported on iOS 16, iPadOS 16.1, macOS 14, tvOS 16 or later, on iPhone, iPad and Apple TV with an A11 Bionic chip or later and Macs with Apple silicon. With the ACME payload, the OS generates a hardware-bound key in the Secure Enclave and the CA can confirm the device is genuine, specific, managed by your MDM and has the expected properties. Erasing or restoring the device removes the key, so the profile must be reapplied.
- IETF device-attest-01: draft-ietf-acme-device-attest-10 adds permanent-identifier and hardware-module identifier types and a device-attest-01 challenge that carries a WebAuthn attestation object. As checked on 14 November 2026 it is in the RFC Editor queue with intended status Proposed Standard; details may change before publication.
- Zero-touch for network devices: BRSKI (RFC 8995) uses the manufacturer identity and a voucher to let the owner's registrar accept a new device, then hands over to EST.
Workloads: when none of the three fits
ACME, EST and SCEP were designed for long-lived subjects with stable names. Containers and functions that live for minutes, scale to thousands of instances and need identity for mTLS fit a different model. SPIFFE describes SVIDs as short-lived cryptographic identity documents that workloads obtain through a Workload API, with implementations such as SPIRE handling automatic rotation and issuing on the basis of node and workload attestation. Use ACME for the ingress certificate that faces clients, and a workload identity system for service-to-service identity behind it. See the mTLS Guide and Cloud Workload Identity Guide.
Decision guidance by organisation size
- Small organisations: use a public ACME CA for internet-facing names and the MDM's built-in certificate service for devices.
- Mid-sized organisations: add a private CA with an ACME interface for internal TLS, keep MDM SCEP with validated challenges for laptops and phones, and plan attested ACME for Apple fleets on supported versions.
- Large and regulated organisations: run all three deliberately, each behind an RA with explicit name policy, plus a workload identity platform. Record which protocol serves which population in the certificate inventory so that a protocol or CA incident can be scoped quickly.
Trade-offs: ACME has the broadest server client support and the strongest name validation, but needs names and reachable validation paths. EST is the strongest fit for authenticated device re-enrolment, but client support on endpoints is narrower and bootstrap identity must exist first. SCEP has near-universal MDM and device support, but its security depends on components outside the protocol.
Migrating to the right protocol
- Inventory by enrolment path. For every issuing CA, list which protocol and endpoint issued each certificate population, and the authorisation method used (challenge type, client certificate, account).
- Fix the worst SCEP first. Replace static or absent challenges with MDM-generated dynamic challenges and a validating policy module; take any open SCEP endpoint off untrusted networks.
- Move named servers to ACME. Start with a pilot group of internal web servers against a private ACME CA in a non-production environment, then production, keeping the old issuance path available until two renewal cycles succeed.
- Move capable network devices to EST. Enable EST with certificate-based client authentication on platforms that support it, using explicit trust anchors.
- Pilot attested ACME on Apple devices. Deploy the ACME payload to a pilot ring alongside the existing SCEP profile, confirm Wi-Fi, VPN and access policies accept the new certificates, then retire SCEP for that ring.
- Roll back cleanly. Keep the previous profile or issuer configured but unassigned for one validity period, so a failed rollout can be reversed without re-enrolling every device by hand.
The Certificate lifecycle manager migration playbook covers platform moves in more depth.
Worked scenario
A university runs a public ACME CA for its websites, a static-challenge SCEP endpoint inherited from an old wireless project for staff laptops and lab equipment, and manual certificates on 400 network switches.
- Risk found: the SCEP challenge appears in a published Wi-Fi setup guide, and the template accepts any requested subject, so anyone on campus could request a certificate the RADIUS servers accept.
- Decision: staff laptops move to MDM SCEP with per-device signed challenges; managed Macs and iPads on supported versions move to the ACME payload with attestation; lab equipment that supports only SCEP gets a separate endpoint and template restricted to a lab subject namespace on an isolated VLAN.
- Switches: models that support EST enrol with their manufacturer certificate; the rest stay manual but enter the certificate inventory with renewal alerts.
Where identity fits
Every enrolment path creates non-human credentials that can mint certificates: ACME account keys and DNS API tokens, EAB MAC keys, NDES service accounts, RA certificates and EST bootstrap credentials. Each needs an owner, least privilege, vault storage and rotation, because whoever holds it can issue trusted identities. See the Machine Identity, PKI and Certificate Lifecycle Guide.
Detection and measurement
- Enrolment path coverage: share of certificate populations with a known protocol, endpoint and authorisation method in the inventory.
- Static challenge count: SCEP profiles or endpoints still using reusable challenges (target zero).
- Name policy rejections: requests refused by RA or policy module for subject or SAN mismatch; a sudden rise suggests misuse or misconfiguration.
- Automated renewal rate: share of certificates renewed without human action, by protocol.
- Unexpected issuance: certificates for your names found through Certificate Transparency or CA logs that no known client requested.
- Attested share of device certificates: proportion issued with hardware attestation where the platform supports it.
Common failure patterns
- Choosing by tool default: the MDM offered SCEP, so network devices got SCEP too, with a shared challenge, when they could have used EST with a device certificate.
- Treating SCEP as authentication: assuming the challenge proves who the device is, when it only proves someone saw the profile.
- One DNS token for all ACME clients: zone-wide write credentials copied to every server.
- Enabling EST server key generation for convenience: private keys transmitted across the network for devices that could generate their own.
- Using device protocols for workloads: issuing long-lived certificates to ephemeral containers because an enrolment endpoint already existed.
Framework mapping
| Framework | Reference | Relevance |
|---|---|---|
| NIST SP 800-53 Rev 5 | SC-17 Public Key Infrastructure Certificates | Issuing certificates under an approved policy, which the enrolment path enforces |
| NIST SP 800-53 Rev 5 | IA-3 Device Identification and Authentication | Device certificates from SCEP, EST and attested ACME |
| NIST SP 800-53 Rev 5 | IA-5(2) Public Key-based Authentication | Binding identity to the private key and validating certification paths |
| NIST CSF 2.0 | PR.AA-01; PR.AA-03 | Managing identities and credentials for devices and services; authenticating them |
| CIS Controls v8.1 | Control 1 Inventory and Control of Enterprise Assets; Control 4 Secure Configuration | Knowing which devices enrol, and configuring enrolment endpoints securely |
| ISO/IEC 27001:2022 Annex A | 8.24 Use of cryptography; 5.17 Authentication information | Key management and protection of enrolment secrets |
Decision checklist
- Does the subject have a DNS name it can prove control of? If yes, start with ACME.
- Does the device already hold a manufacturer or bootstrap certificate? If yes, prefer EST.
- Is an MDM the enrolment authority, and does it issue signed, per-device challenges? Only then use SCEP.
- Does the platform support hardware-bound keys with attestation? If yes, evaluate attested ACME.
- Is the subject an ephemeral workload? If yes, use a workload identity system instead.
- Who checks that the requester is entitled to the requested name, and where is that logged?
Practitioner checklist
- Record the enrolment protocol and authorisation method for every certificate population.
- Eliminate static and absent SCEP challenges; require a validating policy module.
- Keep NDES and other enrolment services off the issuing CA and off untrusted networks.
- Scope DNS-01 credentials to delegated validation zones and use one ACME account per environment.
- Use certificate-based client authentication and explicit trust anchors for EST.
- Leave EST server-side key generation disabled unless a device cannot generate keys.
- Pilot attested ACME for supported Apple devices alongside existing SCEP profiles.
- Issue workload identities through a workload identity system, not device enrolment endpoints.
- Monitor Certificate Transparency and CA logs for issuance nobody requested.
- Decommission retired enrolment endpoints and revoke their RA certificates.
Standards and references
- IETF: RFC 8555 Automatic Certificate Management Environment (ACME)
- IETF: RFC 7030 Enrollment over Secure Transport
- IETF: RFC 8894 Simple Certificate Enrolment Protocol
- IETF: RFC 9773 ACME Renewal Information (ARI) Extension
- IETF: Automated Certificate Management Environment (ACME) Device Attestation Extension (Internet-Draft)
- Apple: Managed Device Attestation
- Microsoft Learn: Configure infrastructure to support SCEP with Intune
- SPIFFE: Overview
Related NHI Mgmt Group resources: cert-manager security guide, Enterprise PKI Design and CA Hierarchy Guide, Certificate Expiry Outages and 47-Day TLS Certificates Guide, 802.1X and EAP-TLS Enterprise Wi-Fi Guide.