NCSC, BSI, and NIST ask different things of passkeys, most sharply on recovery, so the decision worth writing down is which sign-in methods and recovery path each class of account may use.
Two national cyber authorities favor the same credential and land in different places. In April 2026 the UK's NCSC said it "will begin recommending passkeys wherever a service supports them, and two‑step verification (2SV) where it does not". Germany's BSI calls passkeys a very secure way to authenticate, and in the same passage says they are no prerequisite under the BSIG, Germany's NIS-2 transposition law. They answer different questions, and NIST adds further answers on emailed codes, synced keys, and recovery.
These texts can't be compared until their reach is stated. They bind differently or not at all.
The NCSC text is a blog assessment. Dave Chismon, the NCSC's CTO for Architecture, published it on 23 April 2026. The page is tagged for cyber security professionals, large organizations, and the public sector. Its comparison is framed for someone else: "from an individual user's perspective", and its conclusion opens with "For individuals logging into websites and apps". The blog says nothing about device-bound passkeys or enterprise rollouts. It links an underlying paper, and this reading covers the blog only. The shift itself "will be reflected through our ongoing refresh of guidance".
The BSI's NIS-2 MFA page reads German law. The duty it describes is the use of "Lösungen zur Multi-Faktor-Authentifizierung oder kontinuierlichen Authentifizierung", which sits in § 30 Absatz 2 Satz 2 Nummer 10 BSIG. (The BSI page prints "Numme. 10", a typo the statute resolves.) Its note on passkeys calls them "eine sehr sichere Lösung der Authentisierung", a very secure solution for authentication, and then says:
"Nichtsdestotrotz werden Passkeys aktuell weder in der NIS-2 Richtlinie noch im BSIG - als nationales Umsetzungsgesetz in Deutschland – im Rahmen der Risikomanagementmaßnahmen erwähnt und stellen daher keine Voraussetzung gemäß BSIG dar."
In translation: passkeys are currently named in neither the NIS-2 directive nor the BSIG, the German transposition law, as part of the risk management measures, and so are no prerequisite under the BSIG. BSI adds that nothing speaks against organizations engaging with them already. What the page itself recommends is MFA using "geeigneten Wissens- und Besitzfaktoren", suitable knowledge and possession factors. The page also says the law sets no trust level for the method used: "Auf Basis der verpflichtenden Risikoanalyse muss die betroffene Einrichtung selbst entscheiden". The entity decides, on the basis of its mandatory risk analysis. The page is undated, and this reading dates from 3 November 2026. It speaks to German law only, so it can't stand in for another member state's transposition.
BSI TR-03188 is a graded technical guideline. Version 1.0 is dated 15 June 2026. BSI describes it as "Sicherheits- und Handlungsempfehlungen", security and action recommendations, for "alle, die eine Webseite betreiben oder einen Online-Dienst anbieten". Inside a trust level it speaks in MUSS and DARF NICHT. On my reading, those words bind a party claiming that level and not the wider market. Neither BSI page states a legal duty to follow it. The TR also aims "einen Stand der Technik zu definieren", to define a state of the art, and § 30 BSIG uses that phrase too. Whether one feeds the other is a legal question that none of these texts answers.
NIST SP 800-63B-4 is US federal guidance. It "focuses on the authentication of subjects who interact with government information systems", and it is addressed to credential service providers. Its abstract adds that it is "not intended to constrain the development or use of standards outside of this purpose".
None of the four is Swiss, and none names Swiss companies. If you sit outside their reach, you can read them as reasoning to borrow. Nothing below says what a Swiss company must do.
NCSC and NIST weaken the case for emailed codes on different grounds. BSI's two texts don't name them as a sign-in factor. Each claim has to stay the size of its source.
NCSC comes closest to naming them:
"All traditional MFA methods – including passwords combined with SMS codes, email codes, time-based One Time Passwords generated by apps or physical tokens, push approvals – are inherently phishable."
Read the sentence closely. It assesses an email code as one half of password-based MFA, for an individual user. It does not assess a passwordless emailed code. The reasoning carries over, in my view, since a code typed by hand can be relayed whichever factor precedes it. That step is mine, and NCSC doesn't take it.
NIST has a harder sentence with a narrower target. Sec. 3.1.3.1 says:
"Email SHALL NOT be used for out-of-band authentication because it may be vulnerable to:- Access using only a password- Interception in transit or at intermediate mail servers- Rerouting attacks, such as those caused by Domain Name System (DNS) spoofing"
The note directly after it limits the bar:
"Confirmation codes that are sent to validate email addresses or are issued as recovery codes (see Sec. 4.2.1.2) are not authentication processes and not affected by the above prohibition."
So "NIST bans email codes" overstates it. The bar is on email as out-of-band authentication. Does a passwordless emailed code reach any assurance level, then? NIST doesn't name the case, so what follows is my reading. Sec. 2.1.1 lists the types AAL1 "SHALL use". Out-of-band is closed to email by name. A single-factor OTP has "an embedded secret" used as a seed. A look-up secret is a stored record of secrets. A password is "typically chosen by the subscriber". The remaining types are multi-factor OTP and cryptographic authenticators, and a code arriving by email is neither. So on this reading it reaches no AAL. Sec. 3.2.5 settles the phishing question for every code of this shape: authenticators that "involve the manual entry of an authenticator output" SHALL NOT be considered phishing-resistant, "because the manual entry does not bind the authenticator output to the specific session being authenticated". Sec. 2.2.2 separately says verifiers SHALL offer "at least one phishing-resistant authentication option at AAL2".
The BSI's NIS-2 MFA page says nothing on the topic. "SMS" does not occur on it, and "E-Mail" appears only in site navigation and in passages on email encryption. That is silence, and it counts as neither approval nor rejection. TR-03188 comes nearer: its introduction says all common OTP methods are "anfällig gegenüber komplexeren Phishing-Angriffen", open to more complex phishing that uses intercepted data in real time. It does not name email.
The BSI page does set a test any multi-step flow should be read against. A single factor, suitably designed, can reach the level "normal". At the levels "substanziell" and "hoch", it asks for two factors from different categories, and adds:
"Ein mehrstufiger Anmeldeprozess ist ungeeignet."
A multi-step sign-in is unsuitable. The page defines multi-step as factors "nacheinander separat geprüft", checked separately one after the other, and its example is a username and password followed by a one-time code from a TAN generator. An emailed code followed by an authenticator code is also checked in sequence, so on its face it looks like the thing described, and so does ordinary password-then-code 2SV. BSI doesn't discuss that flow, and the law prescribes no level. So this is my reading, and BSI has made no such finding.
NIST and BSI draw nearly the same top line: neither accepts synced keys at its highest level. Sec. 2.3.2 of the NIST text says "syncable authenticators SHALL NOT be used at AAL3". Table 15 of the TR-03188 PDF says of "hoch": "ausschließlich Device Bound Passkeys sind zugelassen", only device-bound passkeys are permitted. Even there, the TR says of two of its attacker models, AM-4 and AM-5, that protection "kann ... aktuell nicht sichergestellt werden": it can't currently be assured.
Below the top line they differ. NIST allows synced keys at AAL2 under Appendix B, which is normative. Its general conditions cover approved cryptography, encrypted storage in the sync fabric, access control, documentation, and a view of what has synced. Most of that falls on whoever runs the sync fabric. As I read it, a relying party can check little of it from its side. One condition is the one to carry away:
"User access to authentication keys in the sync fabric SHALL be protected by AAL2-equivalent MFA to preserve the integrity of the authentication protocols using the synced keys."
NIST's footnote scopes federal enterprise use to staff, contractors, and mission partners and not to public-facing cases. That use adds conditions on FISMA-moderate sync fabrics, device management, and agency-managed accounts, plus a SHOULD on attestation.
NCSC reaches the same place as the MFA condition from the other direction: "The key control is the strength of the authentication protecting the sync account". NCSC sets no level for that account, and NIST sets AAL2-equivalent MFA.
TR-03188 has no row that excludes synced keys at "substantiell". The device-bound rule appears only at "hoch". But no sentence positively accepts them at the lower level either, so that is an inference, and SBT-6 complicates it. It says the authenticator "MUSS eine Form der Attestation bereitstellen" and the relying party must verify it. NIST goes the other way for public services: "The unavailability of attestations SHOULD NOT block the use of syncable authenticators for broad public-facing applications." The TR doesn't say whether a synced passkey can satisfy SBT-6, and I haven't tested it.
The two also ask different things of the verifier on user verification, which decides when a passkey counts as two factors. NCSC gives the principle: "When user verification is required as part of the login, FIDO2 authentication constitutes multi‑factor authentication." NIST's Appendix B has a section addressed to federal agencies acting as relying parties. It says "Verifiers SHALL indicate that UV is preferred" and has them inspect the flag. An unverified result is treated as a single-factor authenticator. TR-03188 SBT-1 says the relying party "MUSS" set userVerification to "required", at both credential creation and assertion, and check the flag.
Side by side, they look like this. The block below is an illustrative summary and not a working configuration. The keys are my shorthand. TR-03188 uses userVerification itself, and NIST writes "UV".
# Illustrative only: what each text asks a verifier to do with a passkey.
nist_800_63b_appendix_b: # AAL2; addressed to federal relying parties
userVerification: preferred # an unverified result still authenticates, as one factor
uv_flag: inspect
backup_eligible: may_restrict # verifiers MAY set a policy on syncable keys
backup_state: may_restrict # agencies SHOULD NOT do this for public-facing apps
attestation: not_a_blocker # SHOULD NOT block broad public-facing use
tr_03188_substantiell:
userVerification: required # SBT-1: no single-factor outcome to handle
uv_flag: check
attestation: verify # SBT-6
tr_03188_hoch:
device_bound_only: true # Table 15, SG-21: closes the sync path entirelyA passkey's phishing resistance holds only as far as the recovery path. The TR names the attack as its model AM-3: a relying party whose recovery process sits at a lower trust level than its FIDO authentication. "Ein Angreifer kann diesen Prozess ausnutzen, um die FIDO-Authentisierung zu umgehen": an attacker can use that process to get around the FIDO authentication. So SG-19 asks that recovery happen at the same trust level or a higher one.
SBT-4 turns that into a line about email:
"Die Relying Party MUSS Account-Wiederherstellung über ein Verfahren anbieten, das einer Authentisierung auf dem Vertrauensniveau „substantiell“ oder höher entsprechen. Insbesondere DARF Passwort- oder E-Mail-basierte Wiederherstellung NICHT angeboten werden."
In translation: the relying party must offer account recovery through a procedure equivalent to authentication at "substantiell" or above, and in particular password- or email-based recovery must not be offered.
NIST takes a different route. Its Sec. 3.1.3.1 note says issued recovery codes "are not authentication processes". Sec. 4.2.1.2 then lets the subscriber keep a recovery address, "postal, email, text message, voice", and gives a code sent to email a 24-hour validity. The distinction runs through the concept itself: BSI asks recovery to match authentication, and NIST places it in a separate category with its own rules.
| NIST SP 800-63B-4 Sec. 4.2 | BSI TR-03188 at "substantiell" | |
|---|---|---|
| Addressed to | Credential service providers, under US government guidance | Passkey relying parties claiming a BSI trust level |
| Recovery code sent by email | Permitted, alone only for an account without identity proofing and otherwise in stated combinations; valid 24 hours | "DARF ... NICHT angeboten werden" (SBT-4) |
| Standard for recovery | Its own rules by account class | Equivalent to authentication at "substantiell" or above (SBT-4, SG-19) |
| Account at AAL2 | Two codes by different methods, or one code plus a bound single-factor authenticator, or repeated identity proofing | Uses trust levels, not AALs |
| Account never identity-proofed | Saved code, issued code, or recovery contact; reproofing is unavailable | Not addressed in the clauses read |
| Notice to the subscriber | "SHALL cause a notification" | Not addressed in the clauses read |
The scopes differ, so this table compares two answers to one engineering question, and it shows no conflict of law. SBT-4 addresses recovery at a passkey relying party. It can't be restated as a BSI ban on emailed sign-in codes. NIST's permission is also narrower than it looks. For an account that can authenticate at AAL2, one emailed code isn't enough.
NCSC stays out of this. It says services "should give users clear ways to manage and remove credentials and to set up recovery options", which asks that recovery exist and sets no bar on its channel.
The decision worth making is per account class. A single sign-in policy forces every account to the level of the most sensitive one, or leaves the most sensitive one at the level of the rest. The table below is my judgment, built on the clauses above. It is not a compliance mapping, and the classes are generic.
| Account class | What I'd allow | What I'd keep out | The clauses behind it |
|---|---|---|---|
| Everyday user | Passkey offered and preferred, synced keys accepted, traditional 2SV where no passkey exists | Counting an emailed code alone as two factors | NCSC recommendation; NIST Sec. 2.2.2 and Appendix B; BSI page on two factors |
| Administrator | Passkey with user verification required, synced keys only under a written rule that the sync account has AAL2-equivalent MFA, which a relying party can check little of | Any manually entered code as the only way in | TR-03188 SBT-1; NIST Sec. 3.2.5 and Appendix B; NCSC on the sync account |
| Highest-assurance account | Device-bound passkeys | Synced keys | NIST Sec. 2.3.2; TR-03188 Table 15 and SG-21 |
| Recovery path, for each class above | A route as strong as the class it recovers | An emailed code as the single step, at least for administrators | TR-03188 SG-19 and SBT-4; NIST recovery rules |
The last row is where I'd spend the time. A saved recovery code plus a recovery contact is one NIST AAL2 combination. Whether that combination counts as "substantiell" I don't know yet, and I don't read SBT-4 as settling it. If you can write down only one thing, write down the recovery path. It's the row that can undo the others.
You don't need a passkey rollout for every class at once. Write one line per class of account. Name the sign-in methods it may use and the path it recovers through, and mark every place where the two don't match. Those lines already show you which exceptions you are carrying.
Qualified counsel decides what a given regime requires of you. The written lines tell them what you actually do.
devguard is the compliance platform I work on, and its changelog lists releases, including sign-in with an emailed code, a passkey, or single sign-on.
Vadim leads engineering, the architecture, automation, and integrations that let compliance live. He's set on making a serious, audit-grade system genuinely easy to run.
Start free, bring your frameworks, and keep the evidence attached as the work ships. No setup friction, no audit-week fire drill.
Start free, bring your frameworks, and keep the evidence attached as the work ships. No setup friction, no audit-week fire drill.