Most IoT fleets are still onboarded with a shared password, a shared claim certificate or a technician typing credentials into a web page, and that first step decides how trustworthy every device will be for the rest of its life. Onboarding is the moment a device proves where it came from, learns who its owner is and receives the network and application credentials it will use from then on. Done badly, it hands an attacker a fleet-wide secret or lets a counterfeit device join the network as a trusted one. This guide compares the main zero-touch onboarding protocols (FIDO Device Onboard, BRSKI, Wi-Fi Easy Connect and cloud provisioning services), explains how to choose between them and gives a phased rollout, monitoring metrics and the failure patterns we see most often. It is written for IoT platform engineers, OT and network architects and the security teams that sign off device programmes. For device identity across the whole lifecycle, see the Identity for IoT Fleets Guide; this page concentrates on the onboarding step itself.
Key takeaways
- Onboarding is two jobs, not one. Network-layer onboarding gives a device the credentials to join a network; application-layer onboarding binds it to a management platform and owner. NIST SP 1800-36 treats them as separate, complementary capabilities, and most programmes need both.
- Start from a factory identity. Every serious zero-touch protocol relies on a key provisioned at manufacture (an IEEE 802.1AR IDevID, an FDO device attestation key, a TPM endorsement key or a Matter-style device attestation certificate). Without one, "zero touch" usually means "zero authentication".
- Ownership must be transferable and provable. FDO ownership vouchers and BRSKI vouchers (RFC 8366) let a buyer prove it owns a device without the manufacturer knowing the final customer at build time. Treat voucher signing keys as high-value machine credentials.
- Shared bootstrap secrets are a fleet-wide single point of failure. A claim certificate or group symmetric key embedded in every unit means one extracted secret can enrol rogue devices; constrain, time-limit or eliminate it.
- Replace bootstrap credentials after onboarding. Once the device is owned, rotate to operational credentials (an LDevID or owner-issued certificate), disable the onboarding path and record the result in the asset inventory.
What device onboarding has to achieve
Whatever protocol you use, a secure onboarding flow has to answer five questions in order. If any one is skipped, the later steps inherit the gap.
- Is this a genuine device? The device proves possession of a manufacturer-provisioned key, ideally held in a secure element or TPM, and the verifier checks it against a manufacturer trust anchor.
- Does this device belong to us? An ownership artefact (FDO ownership voucher, BRSKI voucher, a purchase-order record or an enrolment list) ties the device serial or GUID to the organisation that is onboarding it.
- Is the device talking to the real owner? The device must authenticate the owner or registrar too, otherwise an attacker on the same network can claim it first. Mutual authentication is what separates real zero touch from "first one to connect wins".
- What credentials and configuration does it need? Network credentials (Wi-Fi, 802.1X certificate, Thread network key), platform credentials (an MQTT client certificate) and initial configuration such as endpoints and policies.
- What happens to the bootstrap material? Onboarding credentials are retired or locked, operational credentials are recorded, and the device appears in the inventory with its owner, location and firmware version.
NIST frames network-layer onboarding as provisioning a device with the credentials it needs to connect to a network securely, and recommends combining it with device attestation, application-layer onboarding, lifecycle management and device intent enforcement such as Manufacturer Usage Description (MUD, RFC 8520) profiles. See the Device and IoT Identity Guide for the identity model behind these steps.
Zero-touch onboarding protocols compared
The protocols below solve overlapping problems at different layers. Many fleets combine two of them, for example Wi-Fi Easy Connect to join the network and FDO or a cloud provisioning service to bind to the platform.
| Protocol | Layer | Device trust anchor | Ownership proof | Best fit |
|---|---|---|---|---|
| FIDO Device Onboard (FDO) | Application (platform binding) | Device attestation key (ECDSA P-256/P-384 or Intel EPID) set at manufacture | Ownership voucher: a chain of signed public keys extended at each sale | Edge gateways, industrial PCs, servers and IoT devices sold through distribution where the final owner is unknown at build time |
| BRSKI (RFC 8995) | Network bootstrap, then EST enrolment | IEEE 802.1AR IDevID certificate | Voucher signed by the manufacturer's MASA (RFC 8366), plus MASA audit log | Enterprise and ISP network equipment, industrial switches, devices on IPv6 networks with a registrar |
| Wi-Fi Easy Connect (Device Provisioning Protocol) | Network (Wi-Fi) | Device bootstrapping public key, shared by QR code, NFC or cloud download | Possession of the bootstrapping key (scanned or delivered) | Headless Wi-Fi devices in buildings, retail and homes |
| Matter commissioning | Application and network (Thread or Wi-Fi) | Device attestation certificate chained to a product attestation authority | Setup code or QR code held by the installer | Smart building and consumer devices in the Matter ecosystem |
| Cloud provisioning services (for example Azure IoT Hub DPS, AWS IoT fleet provisioning) | Application (cloud platform) | X.509 certificate, TPM endorsement key or symmetric key | Enrolment list or group, CA registration, or claim certificate | Fleets that connect straight to one hyperscaler IoT platform |
FIDO Device Onboard in more detail
FDO is specified by the FIDO Alliance. The v1.1 specification (Proposed Standard, dated 19 April 2022) defines four sub-protocols, and the Alliance now lists v2.0 as the current version, so check which version your platform implements.
- Device Initialize (DI): at manufacture, loads the onboarding credentials and creates the initial ownership voucher. The device stores a hash of the manufacturer's public key and an HMAC secret that later detects reprogramming outside the factory.
- Transfer Ownership 0 (TO0): the owner registers its address with a Rendezvous Server, keyed by the device GUID.
- Transfer Ownership 1 (TO1): on first power-up the device contacts the Rendezvous Server and learns where its owner is. The Rendezvous Server is not trusted to authorise the transfer, only to point the way.
- Transfer Ownership 2 (TO2): device and owner authenticate each other, the device verifies the ownership voucher chain back to its stored manufacturer key hash, and the owner delivers configuration through ServiceInfo modules. After TO2 every FDO key and identifier is replaced except the device attestation key.
The ownership voucher is the security core. Each entry is signed by the previous owner and names the next, so a manufacturer can sign to a distributor, the distributor to a reseller and the reseller to the end customer. The spec uses EAT tokens signed with COSE_Sign1 for attestation, ECDH or finite-field key exchange in TO2 and AES-GCM or AES-CCM for the session. Rendezvous Servers and owner services must support HTTP and HTTPS; devices must support one of them. RendezvousInfo can list a local Rendezvous Server before an Internet one, which matters for closed OT networks. The FIDO Alliance runs a certification programme covering FDO devices, onboarding services and Rendezvous Servers.
BRSKI in more detail
BRSKI (Bootstrapping Remote Secure Key Infrastructure, RFC 8995, Proposed Standard, May 2021) automates bootstrap for network devices. The new device (the pledge) presents its IDevID in a TLS handshake to a registrar, usually through a join proxy on the link-local network, and sends a signed voucher-request containing a nonce. The registrar forwards it to the manufacturer's MASA, receives a signed voucher and passes it back. The pledge verifies the voucher, adopts the registrar's trust anchor (it "imprints") and can then enrol for an LDevID using EST (RFC 7030) over the same session, under /.well-known/brski. The MASA keeps an anonymised audit log of previous owners that the registrar can review before accepting a device. Nonceless vouchers can be requested in advance for sites that cannot reach the MASA, at the cost of weaker freshness. RFC 8995 states that BRSKI is not intended for constrained low-power lossy networks. See the IoT and Device Certificate Provisioning (EST, SCEP) Guide for the enrolment step.
Cloud provisioning services
- Azure IoT Hub Device Provisioning Service: supports individual enrolments (X.509 leaf certificate or TPM) and enrolment groups (X.509 or symmetric key). Devices present an immutable ID scope to the global endpoint, and allocation policies (evenly weighted, lowest latency, static, or a custom Azure Function) choose the target hub. Microsoft strongly recommends storing device secrets in an HSM.
- AWS IoT Core: offers just-in-time provisioning or registration when a unique certificate is installed before delivery, provisioning by trusted user (a temporary claim valid for five minutes, created with
CreateProvisioningClaim) and fleet provisioning by claim, where every device ships with the same claim certificate. AWS notes that a compromised claim certificate exposes the fleet's onboarding path and that deactivating it does not affect devices already provisioned.
Choosing an approach
| Situation | Recommended pattern | Trade-off |
|---|---|---|
| You control manufacturing and can inject a unique key per unit | Unique X.509 or IDevID per device, registered CA, just-in-time provisioning or BRSKI | Needs a manufacturing PKI and secure key injection at the factory or a secure element supplier |
| Devices pass through distributors and the end owner is unknown at build time | FDO with ownership voucher extension at each sale | Voucher handling must be built into the supply chain and ERP process |
| Headless Wi-Fi devices installed by technicians | Wi-Fi Easy Connect for network join, then platform binding with certificates | Bootstrapping keys on labels must be protected from photography and resale |
| Legacy devices with no secure storage | Trusted-user or technician-assisted provisioning with short-lived claims, on an isolated staging network | Slower, depends on human process; plan replacement |
| Brownfield fleet already using a shared claim certificate | Restrict the claim policy to provisioning only, add a pre-provisioning check against an allow-list of serials, rotate the claim and move new production to unique certificates | Interim risk remains until the old claim is retired |
Small versus large organisations: a small deployment of a few hundred devices can often use the cloud provider's individual enrolment with unique certificates and a manual allow-list. Large fleets and device makers need a manufacturing PKI, voucher automation and an onboarding service that is itself monitored as critical infrastructure, because it can admit or reject thousands of devices an hour.
Securing the onboarding infrastructure
The onboarding service, the registrar, the MASA and the voucher signing keys are some of the most powerful non-human identities in an IoT programme. Whoever controls them decides which devices are trusted.
- Voucher and CA signing keys: keep manufacturer, MASA and owner signing keys in an HSM, with dual control for issuing or extending vouchers in bulk. Log every signature.
- Claim and group credentials: grant a claim certificate only the permission to call the provisioning API, never data-plane topics. Time-limit it where the platform allows, and keep a revocation runbook ready.
- Enrolment allow-lists: enrol only serials or GUIDs that match purchase records. AWS pre-provisioning hooks and Azure custom allocation functions are natural places to enforce this.
- Service accounts and API keys: the ERP or MES integration that pushes device records into the onboarding service holds a machine credential that can enrol devices. Vault it, scope it and rotate it. See the Identity for IoT Fleets Guide.
- Staging networks: onboard on an isolated VLAN or SSID that only reaches the provisioning endpoints, then move the device to its operational zone. In OT, place the local Rendezvous Server or registrar in a DMZ conduit. See the OT Network Segmentation, Purdue Model and IEC 62443 Zones and Conduits Guide.
- Device-side protection: store the attestation key in a secure element or TPM so it cannot be cloned. See the IoT Device Identity with TPM and DICE Guide.
Implementation roadmap
- Inventory current onboarding paths. List every device model, how it receives credentials today, whether any secret is shared across units and who can enrol devices. Rank models by fleet size and exposure.
- Define the identity chain. Decide the factory identity (IDevID, attestation key, TPM EK), the operational identity (LDevID or platform certificate) and who issues each. Write certificate profiles and lifetimes.
- Build the ownership process. For FDO, agree with suppliers how vouchers are delivered with each shipment and extended to you. For BRSKI, confirm each manufacturer runs a MASA and how you reach it. Tie both to purchase orders.
- Pilot in audit mode. Run the new flow on a staging network with one device model and one site. Log every attempt, accept only allow-listed devices and compare the result with the inventory before going further.
- Expand in rings. Move to one production site, then a region, then the whole estate, with a go or no-go review at each ring based on failure rate and support tickets.
- Retire bootstrap secrets. Rotate or revoke shared claim credentials, close the legacy onboarding path in firmware and record the change.
- Plan rollback and re-onboarding. Keep a documented technician-assisted path for devices that fail, and a factory-reset procedure that returns a device to a known pre-onboarding state without leaving old owner credentials behind. Test resale and RMA flows, where the voucher must be transferred or revoked.
Detection and measurement
Log the onboarding service like an identity provider. Events worth alerting on include enrolment attempts from serials not in the allow-list, repeated attestation failures from one source address, a single claim credential used from many geographies, voucher extensions outside business process, devices onboarded to an unexpected owner or hub, and re-onboarding of a device that is already active.
| Metric | What good looks like |
|---|---|
| Share of new devices onboarded with a unique, hardware-protected factory identity | Rising to 100% for new production; legacy exceptions listed with replacement dates |
| Devices still holding a bootstrap or claim credential after onboarding | Zero, checked by configuration audit |
| Rejected enrolment attempts investigated within SLA | All, with a defined triage owner |
| Onboarded devices reconciled to the asset inventory within 24 hours | Above 99% |
| Mean time to revoke a compromised claim or voucher signing key (from exercise) | Hours, not days |
| First-attempt onboarding success rate | High and stable; sudden drops signal certificate or voucher problems |
Framework mapping
| Framework | Reference | How onboarding supports it |
|---|---|---|
| NIST CSF 2.0 | PR.AA-01, PR.AA-03, ID.AM-01 | Hardware identities and credentials are managed; devices are authenticated; onboarding feeds the hardware inventory |
| NIST SP 800-53 Rev. 5 | IA-3 Device Identification and Authentication; SC-12; SC-17; SR-11 Component Authenticity; CM-8 | Device authentication before connection, key and certificate management, anti-counterfeit checks, component inventory |
| CIS Controls v8.1 | Control 1 Inventory and Control of Enterprise Assets | Every onboarded device is recorded; unknown devices are rejected |
| ISO/IEC 27001:2022 Annex A | A.5.9 Inventory of information and other associated assets; A.8.24 Use of cryptography | Asset registration at onboarding; key management for device and voucher keys |
| MITRE ATT&CK | T1200 Hardware Additions; T1195.003 Compromise Hardware Supply Chain; T1552.004 Private Keys; T1078.001 Default Accounts | Attestation and allow-lists counter rogue or counterfeit devices; hardware key storage counters key theft; unique credentials remove default accounts |
| ETSI EN 303 645 | Provision 5.1 (no universal default passwords) | Unique per-device credentials at onboarding meet the consumer IoT baseline. See the ETSI EN 303 645 Consumer IoT Guide. |
Common failure patterns
- One claim certificate for the whole fleet, with broad permissions. Extracted from one unit's flash, it lets an attacker enrol unlimited rogue devices and sometimes publish to production topics. Scope it to provisioning and plan its retirement.
- Trust on first use without owner authentication. The device accepts whichever server answers first, so anyone on the staging network can claim it. Require the device to verify a voucher or a pinned owner key.
- Bootstrapping codes printed on the box. QR codes and setup codes are photographed in warehouses or sold with second-hand units. Treat them as secrets, and re-key or revoke on resale.
- Bootstrap credentials left active. Onboarding succeeds but the factory credential, default admin account or provisioning endpoint stays enabled, giving a second way in. Default credentials are how botnets spread; see the Mirai and IoT Botnet Families Guide.
- Onboarding not linked to the inventory. Devices join the platform but not the CMDB, so they never receive patching or decommissioning. Make inventory registration a required step of the flow.
- Signing keys held in software. Voucher and CA keys sit on an application server, so one server compromise lets an attacker mint ownership for any device. The MSI Signing Keys Leak 2023 shows what happens when device-trust signing keys escape.
Practitioner checklist
- Require a unique, hardware-protected factory identity (IDevID, FDO attestation key, TPM EK or equivalent) in every new device purchase.
- Choose the ownership mechanism per supply chain: FDO vouchers through distribution, BRSKI MASA vouchers for network gear, enrolment lists for direct purchases.
- Ensure the device authenticates its owner or registrar, not only the reverse.
- Keep voucher, MASA and CA signing keys in an HSM with dual control and full logging.
- Scope any claim or group credential to provisioning only and set a retirement date.
- Onboard on an isolated staging network that reaches only provisioning endpoints.
- Rotate to operational credentials after onboarding and disable the bootstrap path.
- Register every onboarded device in the asset inventory automatically.
- Alert on non-allow-listed serials, attestation failures and unexpected owner or hub assignments.
- Test factory reset, resale and RMA so previous owner credentials never survive.
Standards and references
- FIDO Alliance: FIDO Device Onboard Specification 1.1
- FIDO Alliance: FIDO Device Onboard overview and certification
- IETF: RFC 8995 Bootstrapping Remote Secure Key Infrastructure (BRSKI)
- NIST NCCoE: SP 1800-36 Trusted IoT Device Network-Layer Onboarding and Lifecycle Management
- Wi-Fi Alliance: Wi-Fi Easy Connect
- Microsoft Learn: IoT Hub Device Provisioning Service concepts
- AWS: Device provisioning in AWS IoT Core
Related NHI Mgmt Group resources: Identity for IoT Fleets Guide · IoT and Device Certificate Provisioning (EST, SCEP) Guide · IoT Device Identity with TPM and DICE Guide · IoT Security Guide · IoT Secure Firmware Update Guide