# What is FIDO2 / passkeys?

> A passkey is a sign-in key pair made for one website. Your phone, laptop or security key keeps the private key and signs a challenge from the site, which keeps only the public key to check it.

Explainer by Boston Identity, 2026-10-17. 7 min read. https://insights.bostonidentity.com/glossary/fido2

- What it is: standard
- Defined by: W3C, WebAuthn Level 3 (Recommendation, August 2026)
- Who it's for: Teams running staff and customer sign-in

## What is FIDO2 / passkeys?

A passkey is a sign-in credential built on a key pair: a private key and a matching public key. An authenticator creates the pair for one website, at the site's request and with the user's consent [S1 §1]. The authenticator can be built into a phone or laptop, or it can be a separate device such as a USB security key reached over USB, Bluetooth or NFC (near-field communication) [S1 §1]. The site stores the public key. At each later sign-in, the user proves to the site that they still control the private key [S1 §4].

The rules come from the W3C's Web Authentication standard, WebAuthn, which reached Recommendation status at Level 3 in August 2026 [S1]. It defines the browser interface and what it calls the WebAuthn/FIDO2 protocol between the site's server and the authenticator [S1 §1.1]. In its terminology, passkey is another name for a discoverable credential: one the authenticator can find and use even when the site has not first said who the user is [S1 §4]. The spec's consumer example calls passkey sign-in phishing-resistant [S1 §1.2.1]. Each passkey is scoped to one site, and the authenticator signs the full origin of the requesting page into every response, so a response cannot be replayed against a different origin [S1 §5].

Passkeys come in two kinds. A multi-device credential, commonly called a synced passkey, may be backed up to other authenticators, for example by cloud sync [S1 §1.2.1] [S1 §4]. A single-device credential, commonly called a device-bound passkey, is not eligible for backup [S1 §1.2.2] [S1 §4]. The authenticator that creates the passkey makes that choice at creation, and it is permanent for that passkey [S1 §4].

Think of it like this: A passkey works like a signet ring cut for one door. You press it into fresh wax the door hands you, and the door checks the seal against a pattern it keeps, so the ring never leaves your hand. The comparison breaks on copying: a synced passkey can be backed up to your other devices.

Not the same as: Passkey, WebAuthn and FIDO2 name parts of one system. In the W3C text, a passkey is a kind of credential, WebAuthn is the browser interface for making and using one, and WebAuthn/FIDO2 is the protocol that carries the request and the signed reply [S1 §1.1] [S1 §4].

## How does FIDO2 / passkeys work?

The user, the relying party (the website and its server), the browser and the authenticator all take part [S1 §4]. Registration creates the passkey, and authentication uses it at later sign-ins [S1 §1].

[Figure: A passkey is registered once, then used to sign in at later visits, across four parties.]

1. **The site asks for a new passkey.** A script on the site calls the browser's navigator.credentials.create() with a challenge the server produced, the site's name and the user's account details [S1 §1.3.1]. The browser sets the relying party ID, the domain the passkey will belong to, to the page's own domain unless the site names a parent domain [S1 §4]. Browsers offer this interface only on secure connections, such as pages loaded over TLS [S1 §5].
2. **The authenticator makes a key pair.** The browser finds an authenticator and connects to it, and the authenticator asks the user for an authorization gesture [S1 §1.3.1]. That gesture may involve user verification, such as a PIN or a fingerprint, or it may be only a simple test that the user is present, such as a touch [S1 §4]. It then generates a key pair scoped to that site. The private key is bound to that authenticator and is expected never to be exposed to any other party, not even to the authenticator's owner [S1 §4].
3. **The server stores the public key.** The script sends the new public key to the server, along with attestation, which is evidence about the make and properties of the authenticator [S1 §1.3.1]. The server stores the public key and links it to the user's account [S1 §1.3.1]. The spec says the site must keep a credential record, and recommends that it hold the credential ID, the public key, a signature counter and the backup flags [S1 §4].
4. **The site asks for a sign-in.** At a later visit, the script calls navigator.credentials.get() with a new challenge from the server [S1 §1.3.3]. For a passkey, the site may send an empty list of credential IDs, so it does not need to know who the user is before the prompt appears [S1 §4].
5. **The user approves and the authenticator signs.** The authenticator shows the user the accounts it holds for that site and asks for an authorization gesture [S1 §1.3.3]. As at registration, the gesture may be user verification or only a test of user presence [S1 §4]. When the user picks one, the authenticator signs the site's challenge and other collected data into an authentication assertion, a signed proof of the sign-in [S1 §5.1.4]. The full origin of the requesting page is included in what it signs [S1 §5]. The site's script never gets the credential itself [S1 §5].
6. **The server checks the signature.** The script sends the assertion to the server. The server finds the stored public key by its credential ID and verifies the signature; if it is valid, the account linked to that credential is signed in [S1 §1.3.3]. If the server does not recognize the credential ID, the sign-in fails [S1 §1.3.3].

## Why does FIDO2 / passkeys matter now?

In May, Okta and Microsoft both changed their defaults toward passkeys, in generally available releases [BI-FIND-2026-1185](https://insights.bostonidentity.com/ref/BI-FIND-2026-1185). Okta's WebAuthn authenticator now defaults to passkeys for new orgs, and Microsoft extended system-preferred authentication to the first factor in Microsoft-managed tenants [BI-FIND-2026-1185](https://insights.bostonidentity.com/ref/BI-FIND-2026-1185).

We read FIDO's figures for staff sign-in as slower going [BI-FIND-2026-1268](https://insights.bostonidentity.com/ref/BI-FIND-2026-1268). In the FIDO Alliance's own State of Passkeys 2026 survey of 1,400 decision-makers, 68% of organizations have deployed or are deploying passkeys for employee sign-in [BI-FIND-2026-1268](https://insights.bostonidentity.com/ref/BI-FIND-2026-1268). Yet 57% still rely on phishable methods for employees' primary day-to-day sign-in [BI-FIND-2026-1268](https://insights.bostonidentity.com/ref/BI-FIND-2026-1268). These are survey answers, not sign-in logs [BI-FIND-2026-1268](https://insights.bostonidentity.com/ref/BI-FIND-2026-1268).

Cisco Talos describes the UAT-11985 kit, which relays each Google sign-in step as the victim takes it, covering text codes, authenticator apps and phone prompts [BI-FIND-2026-0874](https://insights.bostonidentity.com/ref/BI-FIND-2026-0874). The kit can detect a passkey and draw a passkey prompt, and Talos's post describes no relay step after that prompt [BI-FIND-2026-0874](https://insights.bostonidentity.com/ref/BI-FIND-2026-0874). FIDO and the enterprise agent drafts disagree about agent identity, but both put a passkey at the human end [BI-FIND-2026-0247](https://insights.bostonidentity.com/ref/BI-FIND-2026-0247). An IETF agent-authentication draft says a user who delegates to an agent should sign in with a phishing-resistant method such as a passkey [BI-FIND-2026-0181](https://insights.bostonidentity.com/ref/BI-FIND-2026-0181). FIDO's Megan Shamas says the job is to prove the human with a phishing-resistant credential and never let the agent hold a raw credential [BI-FIND-2026-0247](https://insights.bostonidentity.com/ref/BI-FIND-2026-0247).

## Where does it fit?

- Compared with TOTP: The UAT-11985 phishing kit passes authenticator-app codes through to the real Google sign-in as the victim enters them [BI-FIND-2026-0874](https://insights.bostonidentity.com/ref/BI-FIND-2026-0874). A passkey's signature covers the requesting origin, so it cannot be replayed against a different origin [S1 §5].
- Compared with CIBA: An OAuth cross-device security draft ranks FIDO's cross-device sign-in first and names CIBA as the fallback when FIDO is not available [BI-FIND-2026-1079](https://insights.bostonidentity.com/ref/BI-FIND-2026-1079). The draft says FIDO's flow proves the two devices are close with Bluetooth Low Energy, while CIBA is still open to attackers who know or can guess a user's identifier [BI-FIND-2026-1079](https://insights.bostonidentity.com/ref/BI-FIND-2026-1079).
- Compared with Verifiable digital credential (VDC): A FIDO Alliance white paper gives the two separate jobs: enrol a person with a verifiable digital credential, then accept passkeys for sign-in [BI-FIND-2025-0296](https://insights.bostonidentity.com/ref/BI-FIND-2025-0296). Its reason is personal data: a passkey discloses none at sign-in, while a credential can pass along more than is needed [BI-FIND-2025-0296](https://insights.bostonidentity.com/ref/BI-FIND-2025-0296).
- Compared with SPIFFE: Cecchetti's answer for AI agents is SPIFFE, where the platform that started the agent vouches for it and hands it short-lived credentials [BI-FIND-2026-0254](https://insights.bostonidentity.com/ref/BI-FIND-2026-0254). A passkey needs a person's biometric, PIN or touch, which software cannot supply [BI-FIND-2026-0254](https://insights.bostonidentity.com/ref/BI-FIND-2026-0254).

## What goes wrong?

- The spec leaves a lost passkey to the site: the user reports it, and the server deletes the credential [S1 §1.3.5]. FIDO's Shikiar says knowledge-based recovery is "a backdoor that's open to attackers" while it stays available [BI-FIND-2025-0206](https://insights.bostonidentity.com/ref/BI-FIND-2025-0206). Yubico's Sarginson gives the example of a lost security key with a text-message fallback, and calls recovery "a massive attack vector" [BI-FIND-2025-0286](https://insights.bostonidentity.com/ref/BI-FIND-2025-0286). MIXI says recovering accounts where passkeys are unavailable is still a challenge [BI-FIND-2026-1200](https://insights.bostonidentity.com/ref/BI-FIND-2026-1200).
- User verification tells the site that the same user performed each ceremony; it does not tell the site who that person is [S1 §4]. If several people know a device's PIN or enrol fingerprints on it, they look like one user [S1 §4].
- When a passkey reports user verification for the first time, the site must not rely on it as an authentication factor [S1 §4]. Turning it on should require another factor of equal strength [S1 §4].
- With attestation set to none, the browser usually replaces the attestation statement with an empty one before the site sees it [S1 §5.1.3]. Yubico's Sarginson grants that synced passkeys resist phishing but says they lack attestation of where they come from; Yubico sells hardware keys [BI-FIND-2025-0285](https://insights.bostonidentity.com/ref/BI-FIND-2025-0285).
- An AI agent cannot sign in with a passkey. Cecchetti notes that FIDO requires a biometric, a PIN or a touch to show a person is present, which software cannot supply [BI-FIND-2026-0254](https://insights.bostonidentity.com/ref/BI-FIND-2026-0254).

## What should a security leader ask?

1. Which of our sign-in and recovery paths still fall back to a text-message code or security questions when a passkey is lost?
2. For staff accounts, do we accept synced passkeys or only device-bound ones, and do our servers record and check the backup flags?
3. How many of our employees still use a phishable method for their main daily sign-in?
4. When an AI agent acts for one of our users, where does that person prove who they are, and is it with a passkey?
5. Do we request attestation when a passkey is registered, and what do we do with it?

## Sources

- [S1] [Web Authentication: An API for accessing Public Key Credentials - Level 3](https://www.w3.org/TR/webauthn-3/) (W3C Web Authentication Working Group, W3C Recommendation, 25 August 2026)

## Evidence

- [BI-FIND-2026-1185](https://insights.bostonidentity.com/ref/BI-FIND-2026-1185): Okta and Microsoft both changed defaults toward passkeys in May, both in GA releases. Okta's WebAuthn authenticator now defaults to passkeys for new orgs and to preferred user verification for all orgs, and it was renamed Passkey (FIDO2 WebAuthn) with its own Sign in with a passkey button. Microsoft extended system-preferred authentication to the first factor in Microsoft-managed tenants, so users with passkeys might sign in without a password, with the rollout due to finish by the end of June. Its registration campaigns can now also nudge users to register passkeys. (from "Three vendors put MCP controls in three different places in May, and only Auth0's was GA")
- [BI-FIND-2026-1268](https://insights.bostonidentity.com/ref/BI-FIND-2026-1268): The FIDO Alliance's World Passkey Day release leads with an estimate of 5 billion passkeys in use, which it says it calculated from public data and its own internal deployment data; the release does not show how. We read the workforce numbers in its own State of Passkeys 2026 survey as slower going. Among 1,400 decision-makers at organizations of 500 or more employees, 68% of organizations have deployed or are actively deploying passkeys for employee sign-in, yet 57% of organizations "continue to rely on phishable authentication methods for employees’ primary day-to-day sign-in". Among those not yet fully passwordless, 16% say passwords plus MFA are currently sufficient and 24% are waiting for technologies and standards to mature. Every one of these is an answer to a survey FIDO commissioned, and none comes from sign-in logs. (from "FIDO estimates 5 billion passkeys, but 57% of organizations it surveyed still rely on phishable staff sign-in")
- [BI-FIND-2026-0874](https://insights.bostonidentity.com/ref/BI-FIND-2026-0874): The UAT-11985 kit sits between the victim and the real Google sign-in and passes each step through as it happens: the email address, then the password, then whichever second factor Google asks for. The attacker's server tells the victim's browser which challenge screen to show over a standing WebSocket connection. Its menu covers text and email codes, authenticator apps, phone prompts, backup codes and recovery email, and Talos's own sequence diagram ends with the operator holding the victim's live Google session cookie. The kit also checks whether the account uses a passkey and can draw a passkey prompt, and the post describes no relay step after that prompt. (from "A phishing kit aimed at Taiwan researchers relays each Google sign-in step and gives the attacker the session")
- [BI-FIND-2026-0247](https://insights.bostonidentity.com/ref/BI-FIND-2026-0247): FIDO's public position is that an agent has no identity of its own worth standardizing. Megan Shamas says she does not care who the agent is, only who the person behind it is: prove the human with a phishing-resistant credential, prove the agent acts for that human, and never let the agent hold a raw credential. That is a different starting point from the enterprise work in this corpus, where the IETF agent-authentication draft requires every agent to carry its own identifier and practitioners say an agent must never borrow its user's. The two camps agree on one thing, the passkey at the human end. (from "FIDO says there is no such thing as agent identity, and asks the industry to wait for its spec")
- [BI-FIND-2026-0181](https://insights.bostonidentity.com/ref/BI-FIND-2026-0181): The draft takes its authorization model from OAuth's existing grants. A user delegates through the authorization code grant and SHOULD sign in with a phishing-resistant method such as a passkey; an agent acting for itself can use client credentials or a JWT authorization grant. Where an LLM, Tool or Agent is built from several microservices, the Agent, LLM or Tools can exchange the access token they received for a transaction token, which is bound to one transaction and passed along the internal call chain. For human approval the draft points to CIBA and says "the agent MUST NOT treat local UI confirmation alone as sufficient authorization", then notes that CIBA only accounts for client initiation and maps poorly to confirmation needed mid-execution. A Tool forwarding the access token it got from the Agent is called an anti-pattern in a note, with no MUST NOT. Our briefing on the MCP authorization specification found that spec states its rule against passing tokens on as a MUST NOT. (from "Six companies' agent-auth draft, since adopted by the WIMSE group, keeps credentials out of the model's reach")
- [BI-FIND-2026-1079](https://insights.bostonidentity.com/ref/BI-FIND-2026-1079): The draft's protocol-selection summary ranks three options. FIDO cross-device authentication, the CTAP 2.2 "hybrid transports" flow that lets a phone act as the authenticator for signing in on another device, "provides the best protection against attacks on the unauthenticated channel". It proves the two devices are close with Bluetooth Low Energy advertisements, while an authorization server, the draft notes, "cannot independently measure or enforce proximity on its own." FIDO2/WebAuthn SHOULD be used whenever the devices are capable and the device being signed in to has no suitable FIDO credential of its own. If FIDO is not available, CIBA is the alternative, though it is still open to attackers who know or can guess a user's identifier. The OAuth device authorization grant (RFC 8628) comes last: it is RECOMMENDED only with additional mitigations, and the draft's guidance is to use it only when other protocols are not viable and to avoid it where the protected resources are "sensitive, high value, or business critical." (from "OAuth's cross-device guidance ranks FIDO first, then CIBA backchannel login, and the device-code flow last")
- [BI-FIND-2025-0296](https://insights.bostonidentity.com/ref/BI-FIND-2025-0296): The FIDO Alliance paper gives passkeys and verifiable digital credentials (VDCs) separate jobs. In most cases, it says, relying parties should enrol a person with a VDC, accept passkeys after that, ask for more VDC attributes on a high-risk transaction, and use a VDC to recover a passkey and a passkey to open the VDC. Its case for using VDCs for attributes and passkeys for login rests on personal data: a passkey discloses none at sign-in, while a VDC can pass along more than is needed when a relying party asks and the holder agrees, and a fake identity-verification portal could capture it. Signatures, the editors write, "do not address contextual misuse, highlighting a gap in current standards." (from "FIDO pairs passkeys with digital ID credentials but spells out steps only for moderate assurance")
- [BI-FIND-2026-0254](https://insights.bostonidentity.com/ref/BI-FIND-2026-0254): Passkeys cannot authenticate an agent, by design, so agent authentication falls to workload attestation. FIDO requires a biometric, a PIN or a touch to show a person is present, which is exactly what software cannot supply; Cecchetti's answer is SPIFFE, where you trust the platform that started the agent to vouch for it and hand it short-lived credentials. In April she said the agent protocols had not adopted it: the Model Context Protocol used ordinary OAuth client IDs and OpenClaw a shared secret on the local machine. FIDO's own marketing chief made the same division two months later from the other side, saying its job is the human behind the agent. (from "Two ways to give an AI agent an OAuth identity clashed in April; by June the drafts had partly reconciled them")
- [BI-FIND-2025-0206](https://insights.bostonidentity.com/ref/BI-FIND-2025-0206): Shikiar and host Jim McDonald both put the weak point at account recovery. Shikiar lists removing knowledge-based credentials from account recovery as a FIDO imperative for 2026, next to passkey UX and workforce use: "as long as they're readily accessible for recovery, there is a backdoor that's open to attackers." McDonald suggests identity verification as the way to unlock a lost passkey. Shikiar ties that to FIDO's identity verification certification, which he says supports possession-based onboarding and recovery. Both speak from opinion and give no data. (from "FIDO puts passkeys in use at over three billion on its own estimate, and weighs a certification for digital wallets")
- [BI-FIND-2025-0286](https://insights.bostonidentity.com/ref/BI-FIND-2025-0286): Sarginson says day-to-day use is "probably the least part of the problem these days" in a security-key rollout. The risk he names is recovery: in his example a user drops the key down a drain and the fallback is SMS, and he calls recovery mechanisms "a massive attack vector these days." His rollout advice is phased, with internal evangelists helping first and executives and admins protected before the rest of the staff. He cites no incident or figure for the recovery risk. (from "A Yubico product manager grants that synced passkeys resist phishing and sells hardware keys on attestation")
- [BI-FIND-2026-1200](https://insights.bostonidentity.com/ref/BI-FIND-2026-1200): MIXI added FIDO2 passkeys to its MIXI ID accounts in 2024, after email one-time codes proved weak against real-time phishing, and according to the FIDO Alliance's case study more than 25% of MIXI ID users have registered one. Help-desk queries about late, resent or mistyped codes have fallen, though the study gives no user count, measurement date or size for either result. MIXI held the rollout until Passkey Autofill had matured across the major operating systems, to avoid the "Passkey not found" errors seen in other early deployments. MIXI's Ryo Ito says account recovery "from environments where passkeys are unavailable" is still a challenge. (from "Over a quarter of MIXI ID users registered a passkey; MIXI says recovering accounts without one is still hard")
- [BI-FIND-2025-0285](https://insights.bostonidentity.com/ref/BI-FIND-2025-0285): Nic Sarginson, Yubico's principal product manager for the YubiKey, grants that platform and synced passkeys resist phishing. What he says they lack is "the attestation of where it's coming from": depending on how it was created, he says, a passkey in an Android or iCloud vault could be shared with anyone in the room who has AirDrop, while a YubiKey credential is created and used on the key and needs both the key and its PIN. For the highest assurance level, he says, guidance from NIST and other bodies always recommends an external hardware device, a smart card or a FIDO token. He describes US federal rules as accepting only smart cards and hardware FIDO keys as phishing-resistant, while a UK government advisory body, he says, recommends a FIDO and passkey strategy. (from "A Yubico product manager grants that synced passkeys resist phishing and sells hardware keys on attestation")

---

© 2026 Boston Identity LLC. All rights reserved. Published by Boston Identity LLC.
Quote up to 100 words with the credit "Boston Identity, IAM Insights" and a link to https://insights.bostonidentity.com/glossary/fido2. Claim numbers may be cited freely. Training, fine-tuning and text-and-data-mining use of this page is permitted.
Terms: https://insights.bostonidentity.com/terms · How to cite: https://insights.bostonidentity.com/cite
