Any login that starts on one device and is approved on another, by scanning a QR code, typing a short code or tapping a push notification, depends on the user deciding whether a request they did not create is legitimate, and attackers exploit exactly that gap. RFC 10027, published in August 2026 as Best Current Practice 247, is the IETF's guidance on securing these cross-device flows. It describes the flow patterns, the consent phishing and session phishing attacks against them, seventeen practical mitigations and which protocol to choose. This guide turns that BCP into design decisions for identity architects, product teams building TV, kiosk, IoT or desktop sign-in, and security teams reviewing existing flows. It complements the attack-focused Device Code Phishing Guide.
Key takeaways
- The channel between devices is unauthenticated: a QR code or user code proves nothing about who started the flow, so the authorisation server cannot tell a user's TV from an attacker's laptop without extra signals.
- RFC 10027 makes risk assessment mandatory: implementers must assess risk before deploying a cross-device flow, select mitigations from the BCP and should avoid the flow if risks cannot be mitigated.
- Proximity is the strongest control: FIDO2/WebAuthn hybrid transport uses Bluetooth Low Energy to prove the devices are near each other and should be used whenever the devices can support it.
- The device authorisation grant is the last resort: RFC 10027 calls it the most exposed option and advises against it for sensitive resources unless additional mitigations are in place.
- Never use cross-device flows on one device: the BCP says they should not be used for same-device scenarios, and authorisation servers may block them there.
Terms and flow patterns
RFC 10027 uses two device roles: the Consumption Device, where the user wants the service (a smart TV, kiosk, car, CLI or desktop app), and the Authorization Device, the personal trusted device used to authenticate and approve (usually a phone). It separates cross-device authorisation (granting access to a resource) from cross-device session transfer (moving an existing session to another device).
| Pattern (RFC 10027 section) | How data moves | Example protocol | Main weakness |
|---|---|---|---|
| User-transferred session data (3.1.1) | Consumption Device shows a QR code or user code; user enters or scans it on the Authorization Device | OAuth 2.0 Device Authorization Grant (RFC 8628) | An attacker can start the flow and send the code to the victim |
| Backchannel-transferred session (3.1.2) | Client asks the authorisation server, which pushes an approval request to the user's device | OpenID Connect CIBA | Anyone who knows or guesses the user identifier can trigger requests, including floods |
| User-transferred authorisation data (3.1.3) | Authorization Device receives a code (for example six digits) that the user types on the Consumption Device | Code-based approval schemes | The user can be tricked into reading the code to an attacker |
| Session transfer (3.2.1) | Authenticated device generates a transfer code or QR; the other device redeems it | OpenID4VCI pre-authorised code flow, "continue on phone" features | The user can be persuaded to send the code or QR to an attacker |
The attacks the BCP addresses
- Cross-Device Consent Phishing (CDCP): the attacker starts a legitimate flow on their own device, then changes the context in which the victim sees the code or request (a fake loyalty email, a security warning, a swapped label on a bike rack) so the victim approves it. The attacker's device receives the tokens. Variants exist for each pattern, including floods of push requests.
- Cross-Device Session Phishing (CDSP): the attacker persuades the user to send them a session transfer code or QR, then takes over the session.
- Case studies: RFC 10027 section 4.3 works through nine scenarios, from streaming service codes and fake file-deletion warnings to bulk push approvals with a fake help desk, network-join QR codes and wallet onboarding.
- Out of scope: users who collude with attackers.
In ATT&CK terms the delivery is usually phishing (T1566.002 Spearphishing Link) and the outcome is stolen tokens (T1528 Steal Application Access Token) that are then used directly (T1550.001 Application Access Token). See the QR Code Phishing (Quishing) Guide for QR-specific lures.
The seventeen mitigations
RFC 10027 section 6.1 lists practical mitigations and classifies each as preventing, disrupting or helping recover from an attack. Section 2 says implementers should include proximity among their selected mitigations.
| Mitigation (section) | Type | What it means in practice |
|---|---|---|
| Establish proximity (6.1.1) | Prevent, disrupt | Physical connection, NFC, BLE or UWB, shared network or geolocation; VPNs and GNSS spoofing can defeat the weaker signals |
| Short-lived codes (6.1.2) | Disrupt | Expire QR and user codes quickly to shrink the window |
| One-time or limited-use codes (6.1.3) | Disrupt | Reject reuse, allowing a small number of retries for usability |
| Unique codes (6.1.4) | Disrupt | Detect the same code submitted repeatedly and refuse tokens |
| Content filtering (6.1.5) | Disrupt | Block messages that send the same code to many recipients |
| Detect and remediate (6.1.6) | Recover | Revoke tokens tied to misused codes and notify resource servers via CAEP/SSF |
| Trusted devices (6.1.7) | Prevent | Only managed or registered devices may start the flow |
| Trusted networks (6.1.8) | Prevent | Only start flows from known networks; combine with other controls |
| Limited scopes (6.1.9) | Recover | Issue narrower scopes to tokens obtained through cross-device flows |
| Short-lived tokens (6.1.10) | Recover | Shorter access and refresh token lifetimes limit damage |
| Rate limits (6.1.11) | Prevent, disrupt | Cap codes and requests per client or user; slows scaled attacks |
| Sender-constrained tokens (6.1.12) | Recover | Bind tokens to a device key so they cannot be moved or resold |
| User education (6.1.13) | Prevent | Set expectations about when requests occur; insufficient alone |
| User experience (6.1.14) | Prevent | Show who started the flow, why and what approval means; make decline the default or equally prominent |
| Authenticate then initiate (6.1.15) | Prevent | Require phishing-resistant authentication on the Consumption Device before starting the flow |
| Request initiation verification (6.1.16) | Disrupt | Send a code to the Authorization Device that the user must enter on the Consumption Device |
| Request binding with out-of-band data (6.1.17) | Disrupt | Bind a serial number, code or location into the QR so the user must confirm it |
Choosing the protocol
| Protocol | RFC 10027 view | Use when | Required mitigations |
|---|---|---|---|
| FIDO2/WebAuthn cross-device authentication (hybrid transport, CTAP 2.2) | Strongest; BLE proves proximity; should be used whenever devices are capable | Both devices have BLE and internet; the Consumption Device supports WebAuthn and CTAP 2.2 hybrid | Can be combined with the authorisation code grant and PKCE, avoiding device grant risks |
| CIBA | Less exposed than the device grant, but vulnerable to known or guessable user identifiers | The Consumption Device can obtain a user identifier (call centre, point of sale) | Proximity and trusted devices or networks should be considered; rate limits against request floods |
| OAuth 2.0 Device Authorization Grant (RFC 8628) | Least constrained, most exposed; avoid for sensitive resources | Input-constrained devices with no alternative | Only with additional mitigations: short-lived one-time codes, limited scopes, short-lived and sender-constrained tokens |
| Authorisation code grant on the same device | Cross-device protocols should not be used for same-device scenarios | The user is on the device that needs access | Standard OAuth and OpenID Connect controls |
Passkeys signed in through the hybrid transport are the practical route to proximity for most consumer and workforce use. See the Passwordless and Passkeys Guide and the FIDO2 and WebAuthn Internals Guide.
Design patterns that hold up
- TV and streaming apps: prefer passkey sign-in with hybrid transport where the TV platform supports it. If a device code is unavoidable, scope tokens to the playback service only, expire codes in minutes, and show the device name and location on the phone approval screen.
- Workforce CLI and desktop tools: use the authorisation code grant with PKCE and a local browser rather than the device code flow. Where the identity provider allows it, block the device code flow by policy for users and applications that do not need it.
- Kiosks and shared terminals: start flows only from registered devices (trusted devices) on known networks, and bind the kiosk identifier into the request so the phone shows which kiosk is asking.
- Push approvals (CIBA and similar): show a binding message with the transaction or a number the user must match, rate limit requests per user, and treat repeated declines as an attack signal. See the MFA Fatigue and Bypass Guide.
- Session transfer ("continue on your phone"): generate transfer codes only after strong authentication, make them single use and very short lived, and warn the user never to share them.
- IoT and network onboarding: prefer physical connection or NFC to prove proximity, and bind the device serial number into the QR so the user confirms the device in front of them.
Decision guidance
- Is a second device really needed? If the device that needs access has a browser and keyboard, use the authorisation code grant with PKCE on that device. Cross-device flows exist for input-constrained or shared devices, not for convenience.
- Can both devices do FIDO hybrid? If yes, use it. It is the only option in the BCP whose security does not depend mainly on the user spotting a fake context.
- Does the Consumption Device know who the user is? If a trusted operator or system supplies the user identifier (a bank branch, a call centre), CIBA with a binding message and rate limits is the next choice.
- Is the resource sensitive? If a device grant is the only option and the resource is sensitive (admin access, email, payments), RFC 10027 advises against it. Reduce the resource instead: issue a narrowly scoped token for the TV app, not mailbox access.
- Consumer versus workforce: workforce identity providers can rely on trusted devices and networks and can switch off the device grant for almost everyone. Consumer platforms cannot assume managed devices, so they lean on proximity, short-lived codes, contextual approval screens and limited scopes.
- Small teams: the quickest wins are policy changes at the identity provider (restrict device code flow, shorten code and token lifetimes) and better approval screen text, before any protocol changes.
Where identity fits
Cross-device flows are often how devices and command-line tools get long-lived refresh tokens, which then behave like non-human credentials on shared or unmanaged hardware. Inventory which clients are allowed to use the device grant or CIBA, give them narrow scopes, and include their refresh tokens in revocation and rotation processes. See the OAuth Token Introspection and Revocation Guide.
How to implement
- Inventory flows: list every client using the device authorisation grant, CIBA, QR login or session transfer, with the scopes and token lifetimes each receives.
- Run the risk assessment RFC 10027 requires: for each flow, record the resources at stake, the attack patterns that apply (CDCP, CDSP) and which mitigations you will use.
- Remove unnecessary cross-device use: move same-device and browser-capable clients to the authorisation code grant with PKCE; block the device grant for everyone else by policy.
- Add proximity where possible: adopt passkeys with hybrid transport for the remaining flows that can support it.
- Harden what remains: short-lived single-use codes, rate limits, narrower scopes, short-lived and sender-constrained tokens, and an approval screen that shows device, location and client and defaults to decline.
- Pilot with monitoring: enable the new controls for one client in report mode, review blocked and challenged attempts, then enforce. Rollback is re-enabling the previous policy for that client only.
- Prepare recovery: make sure misused codes lead to token revocation and that resource servers learn of it (CAEP/SSF or introspection).
Detection and measurement
- Signals to log: code issuance and redemption with IP, ASN and device details on both sides; mismatch between the Consumption Device location and the Authorization Device location; codes redeemed repeatedly; bursts of CIBA requests for one user; approvals followed by immediate bulk access.
- Alert on: device code sign-ins for users or apps not on the allowlist, first-time device grant use for privileged users, and approvals where the two devices are in different countries.
- Metrics: number of clients still permitted to use the device grant; share of cross-device sign-ins using FIDO hybrid; median code lifetime; share of cross-device tokens that are sender-constrained; time from detected misuse to token revocation; approval decline rate (a sudden drop can mean users are approving on autopilot).
Framework mapping
| Framework | Reference | Relevance |
|---|---|---|
| IETF | RFC 10027 (BCP 247) sections 2, 6.1, 6.2 | Mandatory risk assessment, mitigations and protocol choice |
| NIST SP 800-53 Rev 5 | IA-2, SC-23 Session Authenticity | Strong authentication and protecting the session established across devices |
| NIST SP 800-53 Rev 5 | AC-12, IA-11 | Short sessions and re-authentication for tokens from cross-device flows |
| NIST CSF 2.0 | PR.AA-03, DE.AE-02 | Authentication and analysis of anomalous cross-device approvals |
| MITRE ATT&CK | T1566.002, T1528, T1550.001 | Phishing delivery, token theft and token reuse |
Common failure patterns
- Device code flow enabled for everyone: the identity provider allows the device grant for every user and app by default, so attackers can run consent phishing against the whole workforce.
- Approval screens without context: the phone says "Approve sign-in?" with no device, location or app name, and approve is the highlighted button.
- Long-lived, reusable codes: codes valid for fifteen minutes or more and redeemable several times give attackers time to phish and resell.
- Full scopes and long refresh tokens: a TV app obtains the same broad, long-lived tokens as a managed laptop.
- Cross-device flow on the same device: a desktop app uses the device grant because it was easier than a redirect URI, widening exposure for no benefit.
- No recovery path: when misuse is spotted, nobody can find and revoke the tokens issued through the abused codes.
Practitioner checklist
- Inventory every cross-device flow, client, scope and token lifetime.
- Document an RFC 10027 risk assessment for each flow and the mitigations chosen.
- Prefer FIDO2/WebAuthn hybrid transport wherever both devices support it.
- Restrict the device authorisation grant by policy to named clients that genuinely need it.
- Use short-lived, single-use codes and rate limit requests per user and client.
- Show device, location and client on approval screens and make decline the default.
- Issue narrower scopes and shorter, sender-constrained tokens to cross-device clients.
- Correlate locations of both devices and alert on mismatches.
- Test that misused codes lead to fast token revocation and resource server notification.
Standards and references
- IETF: RFC 10027 Best Current Practice for Security of Cross-Device Flows
- IETF Datatracker: draft-ietf-oauth-cross-device-security history
- IETF: RFC 8628 OAuth 2.0 Device Authorization Grant
- OpenID Foundation: Client-Initiated Backchannel Authentication (CIBA) Core
- FIDO Alliance: FIDO2 specifications including CTAP 2.2
- IETF: RFC 9700 Best Current Practice for OAuth 2.0 Security
Related NHI Mgmt Group resources: Device Code Phishing Guide · QR Code Phishing (Quishing) Guide · Phishing-Resistant MFA Guide · OAuth Step-Up Authentication Challenge Guide