Passkeys and multi-signal verification reshape identity fraud controls

At a glance

What this is: NHI Mgmt Group's analysis, drawing on Zumigo's blog post: Passkeys, mobile intelligence, and layered verification are converging as organisations try to reduce fraud without adding user friction.

Why it matters: Identity teams need to treat passkeys, device intelligence, and AI-assisted identity checks as connected controls, not separate projects, because fraud now crosses consumer, workforce, and agent-driven journeys.

By the numbers:

• Two thirds of 600 developers in Bitwarden’s 2024 survey had used passkeys to access work applications.
• Almost 90% of those developers had a highly favorable or favorable attitude towards passkeys.
• Nearly one million passkeys were registered every day after Microsoft added support for Xbox and Copilot in 2024.
• Passkeys are expected to account for over 50% of consumer sign-ins in 2026.

Context

Identity verification is moving away from one-time checks and toward continuous signal fusion. Passkeys reduce password risk, but fraud and account takeover now depend on device context, behavioural patterns, and recovery flows as much as on the initial login.

The growing overlap between consumer identity, workforce identity, and agent-assisted activity means teams can no longer govern authentication in isolation. The practical question is how to verify trust when a phone, a device, and an AI-assisted workflow all participate in the same access path.

Key questions

Q: How should security teams roll out passkeys without breaking account recovery?

A: Start with low-risk journeys, then define recovery as a controlled identity workflow rather than a convenience feature. Use step-up verification, help-desk approval, and audit logging for resets. The goal is to keep passwordless sign-in simple while making fallback paths stricter than the primary login path. That is where most account abuse begins.

Q: Why do single-signal identity checks fail against modern fraud patterns?

A: Single-signal checks fail because they rely on one observation, such as a document image or a single device event. Attackers can bypass that with deepfakes, spoofing, SIM swaps, and automated fraud. Security teams need multiple, independent signals that support both possession and ownership, plus verification that continues across sessions instead of stopping at onboarding.

Q: How do AI agent identities change access governance compared with human users and service accounts?

A: AI agents can require task-specific access that is created, used, and retired differently from human or static machine accounts. That means conventional review cycles may miss the moment when privilege is actually exercised. Financial institutions need a governance model that treats agents as distinct identity subjects with runtime access patterns.

Q: What is the best way to reduce consumer account takeover risk?

A: The best way to reduce consumer account takeover risk is to combine strong login controls with fraud detection and step-up verification. Use phishing-resistant MFA where possible, monitor unusual device, location and session behaviour, and trigger step-up checks for risky changes such as password resets, new payees or profile updates. Rate limiting, breached-password checks and secure account recovery close common takeover paths.

Technical breakdown

Passkeys and phishing-resistant authentication

Passkeys replace shared secrets with public-key cryptography bound to the device or authenticator, which removes reusable passwords from the login flow. The security gain comes from origin binding and credential non-exportability, so the verifier checks a cryptographic challenge rather than a memorised secret. That shifts many common phishing and credential stuffing paths, but it also moves risk into enrollment, recovery, and device binding. In practice, the control is only as strong as the lifecycle around it, especially when users can add devices or recover access through weaker channels.

Practical implication: Treat passkey rollout as an authentication lifecycle programme, not a UI change, and harden enrollment and recovery flows accordingly.

Multi-signal identity verification for fraud prevention

Modern identity fraud controls combine phone intelligence, account history, device reputation, document verification, biometrics, and risk scoring because no single signal reliably separates legitimate users from coordinated abuse. The architecture works by fusing weak signals into a confidence decision, often in real time, so the system can approve, step up, or deny access before fraud completes. That approach is useful against organised fraud rings and synthetic identity behaviour, but it also creates governance pressure around false positives, explainability, and appeal paths for legitimate users.

Practical implication: Design fraud checks as a decisioning pipeline with explicit step-up paths and review logic, not as a single binary screen.

Agent identity and personal identity are starting to converge

As users delegate shopping, scheduling, and financial tasks to AI agents, the identity boundary between the human and the software acting on their behalf becomes harder to enforce. The agent may present on the user’s device, reuse the user’s context, and trigger authentication events that look like ordinary consumer activity, even though the runtime decision-maker is different. That creates a new governance problem: who is accountable when the agent acts, what authority it actually has, and how much of the user’s identity context it should inherit. This is where identity governance begins to intersect with agentic AI security in a practical way.

Practical implication: Define delegated authority, user consent, and step-up requirements for agent-assisted workflows before those flows become business critical.

NHI Mgmt Group analysis

Passkeys are becoming a baseline control, but they do not end identity fraud. Passwordless authentication removes a large class of credential theft, yet fraudsters shift to account recovery, device compromise, and social engineering around enrollment. The governance problem moves from secret protection to trust in the binding process. Identity programmes should treat passkeys as necessary but not sufficient, because the weak link becomes the surrounding lifecycle.

Multi-signal verification is replacing single-point identity decisions. Phone intelligence, device intelligence, biometrics, and risk-based authentication are most effective when they operate as a decision mesh rather than isolated checks. That creates a better chance of catching coordinated fraud, but it also increases the need for tuning, auditability, and a defensible exception process. Organisations should expect fraud control to look more like adaptive assurance than static authentication.

Agent-assisted identity flows create a new trust boundary. When an AI system helps a user complete personal or financial tasks, the question is no longer only whether the human authenticated. The issue is how much trust the workflow extends to the software acting in the user’s place. That is an identity governance problem, not just an AI one, and it will need explicit policy for delegated actions, recovery, and revocation.

Verification trust gap: identity systems are being asked to validate users, devices, and software delegates at the same time, but most control designs still assume a single human at the keyboard. That mismatch will drive the next round of fraud controls and access policy changes. Practitioners should rework verification models around who or what is actually acting, not who owns the account on paper.

What this signals

Consumer identity assurance is becoming a continuous control problem. As passkeys spread, the differentiator is no longer whether an organisation supports passwordless sign-in, but how well it can bind that sign-in to a trusted device, a trusted recovery path, and a trusted transaction. Identity teams should expect authentication policy to merge with fraud decisioning.

Agentic workflows will force identity teams to formalise delegation. Once software can shop, pay, or schedule on a user’s behalf, verification models must distinguish between account ownership and action authority. That will push IAM and fraud teams toward explicit policy for delegated actions, consent expiry, and revocation.

For practitioners

• Standardise passkey enrollment and recovery controls Require phishing-resistant enrollment, restrict fallback methods, and review account recovery paths for abuse opportunities before scaling passwordless sign-in.
• Fuse identity signals into a single fraud decision path Combine device reputation, phone intelligence, behavioral risk, and document checks so step-up decisions are made from correlated evidence rather than one score.
• Separate human intent from agent action Document which actions an AI assistant may initiate, what user approval is required, and where revocation or reauthentication must occur.
• Tune exceptions for legitimate users Build review queues and appeal logic for cases where layered verification produces false positives, especially in high-friction consumer journeys.

Key takeaways

• Passwordless authentication lowers exposure to stolen credentials, but it shifts risk toward enrollment, recovery, and device-binding controls.
• Fraud prevention is moving toward layered verification that combines multiple weak signals into a single decision.
• AI-assisted personal workflows will force identity programmes to define delegated authority more explicitly than traditional account-based models do.

Related NHI Mgmt Group resources: Digital Identity, eID and Identity Wallets Guide · Open Source Security Guide · Application Security Statistics Guide

Key terms

• Passkey: A passkey is a passwordless credential based on public key cryptography. A private key stays on the user’s device, while a public key is stored by the service. During login, the device signs a challenge after local unlock, which reduces phishing and eliminates shared secret reuse.
• Risk-Based Authentication: An access model that changes verification requirements based on the estimated risk of the request. It combines identity assurance, device posture, application sensitivity, and contextual signals to decide whether to allow, block, or step up verification before access is granted.
• Biometrics: Biometrics are identity checks that use physical or behavioural traits such as fingerprints, face patterns or voice. They can be targeted through spoofing, replay or poor liveness checks, and they raise privacy and recovery concerns if compromised. Defenders should protect template storage, require strong liveness validation and pair biometrics with additional controls for high-risk access.
• Device Intelligence: Device intelligence is the practice of interpreting signals from a device to assess whether a session or transaction is likely legitimate. It goes beyond fingerprinting by combining device context with behavioural, identity, and payment evidence to support a risk decision.

Deepen your knowledge

The NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, identity lifecycle, and secrets management. It helps practitioners connect identity controls to the access and delegation patterns their programmes actually rely on.