For years, security awareness training told employees to look for spelling mistakes, odd sender addresses and awkward phrasing. Generative AI erased those clues. Attackers can now produce flawless, personalized phishing emails in any language, clone an executive’s voice from a few minutes of public audio, and stage a video call with convincing synthetic participants. Defenses that depend on a human spotting a fake are losing ground.
The good news is that the most effective answer to AI-enabled phishing does not require out-thinking the AI. Phishing-resistant authentication, led by FIDO2 passkeys, removes the shared secret the attacker is trying to obtain. A passkey cannot be typed into a fake page, read out over the phone or approved by a tired user at midnight. This guide explains why legacy MFA fails against modern social engineering, what NIST and CISA now expect, and how to roll out passkeys across a workforce while closing the gaps attackers have moved to: help desks, account recovery and payment approvals.
How AI Changed the Phishing Threat
Phishing has always been a numbers game. Generative AI changes the economics in three ways. First, quality: lures are fluent, contextual and tailored using scraped LinkedIn profiles, press releases and leaked data, so the generic tell-tale signs disappear. Second, scale: a single operator can run thousands of individualized conversations, including real-time chat that responds plausibly to questions. Third, new channels: voice cloning and real-time video manipulation extend impersonation from email into phone calls, voicemail and meetings.
The case security leaders cite most often remains the one reported by CNN in May 2024: an employee in the Hong Kong office of the engineering firm Arup was persuaded, after a video conference in which the other participants appeared to be senior colleagues, to make transfers totaling roughly US$25 million. Every face and voice on the call was fake. The attack did not compromise a single system; it compromised a human decision.
Credential theft has evolved in parallel. Adversary-in-the-middle phishing kits proxy the real login page, relaying the user’s password and one-time code to the genuine service in real time and stealing the resulting session cookie. From the user’s point of view, the login looks normal and even succeeds. Any authentication method that relies on a code or approval the user can be tricked into handing over is vulnerable to this pattern.
Why Legacy MFA Is No Longer Enough
Multi-factor authentication is still far better than passwords alone, but not all MFA is equal. Several popular methods share a structural weakness: the secret or approval can be redirected to an attacker.
- SMS and voice codes can be intercepted through SIM swapping or simply phished on a fake page and replayed.
- Authenticator app one-time codes are phishable in exactly the same way, because the user types them wherever asked.
- Push notifications are vulnerable to MFA fatigue, where attackers trigger repeated prompts until a user approves one. Number matching helps but does not stop a user who is being coached by a convincing caller.
- Knowledge-based recovery questions are trivially defeated by data breaches, social media and AI-assisted research.
The joint FBI and CISA advisory on the Scattered Spider group, published in November 2023, documented exactly these techniques: impersonating IT staff to have passwords and MFA reset, MFA bombing, SIM swapping, and then registering the attacker’s own MFA device to persist. The advisory’s top mitigations included enforcing phishing-resistant MFA. AI does not create a new class of attack here; it makes these social engineering steps faster, cheaper and more convincing.
How Passkeys Deliver Phishing Resistance
A passkey is a FIDO2/WebAuthn credential: a public-private key pair created for one specific website or application. The private key stays on the user’s device or in a secure credential manager, protected by a local unlock such as a fingerprint, face scan or device PIN. During sign-in, the service sends a challenge, the authenticator signs it with the private key, and the service verifies the signature with the stored public key.
Three properties make this resistant to phishing, including AI-enhanced phishing:
- Origin binding. The browser or operating system only offers a passkey to the exact domain it was registered for. A look-alike domain, however perfect, simply receives nothing.
- No shared secret. The server holds only a public key. There is nothing for the user to type, read aloud or forward, and nothing useful to steal from a server breach.
- User presence and verification. The authenticator requires a deliberate local action, which defeats remote prompt bombing.
Passkeys come in two broad forms. Synced passkeys are stored in a platform or third-party credential manager and synchronized across a user’s devices, which makes them convenient and resilient to device loss. Device-bound passkeys live on a single authenticator, such as a hardware security key or a managed device’s secure element, and cannot be exported. NIST’s revised digital identity guidelines, SP 800-63-4, finalized in 2025, explicitly recognize syncable authenticators under defined conditions at authentication assurance level 2, while hardware-bound, non-exportable authenticators are associated with the higher AAL3. Many enterprises use both: device-bound keys for administrators and high-risk roles, synced passkeys for the broader workforce and customers.
Where Adoption Stands in 2026
Passkeys have moved from early adopter to mainstream. According to the FIDO Alliance’s State of Passkeys 2026 report, as summarized by Descope, consumer awareness reached about 90 percent, and 68 percent of surveyed organizations were deploying, piloting or rolling out passkeys for employee sign-in. The same research shows the gap that remains: only around 30 percent use passkeys as the primary workforce sign-in method, while a majority still rely mainly on passwords or other phishable methods. The top barriers organizations cited were legacy system compatibility, budget approval and concerns about device recovery, while user resistance ranked low.
Regulators and platform vendors are pushing in the same direction. Federal guidance in the United States has required phishing-resistant MFA for agency systems since OMB M-22-09, CISA promotes it as the gold standard, and major identity providers now make passkey registration a default option for workforce accounts. The practical question for most organizations is no longer whether to adopt passkeys, but how to get from a pilot to passkeys being the default and phishable fallbacks being retired.
A Rollout Playbook for the Workforce
Successful passkey programs treat the project as a change to identity architecture, not a feature toggle. The following sequence works for most mid-size and large organizations.
1. Map Identities, Applications and Authenticators
Inventory where users authenticate: your primary identity provider, applications federated through SSO, legacy applications with their own logins, VPN and remote access, privileged access tools and service desks. Note which already support FIDO2/WebAuthn directly or through the identity provider, and which will need a proxy, an upgrade or replacement. This is usually where the legacy compatibility barrier becomes concrete and fundable.
2. Choose Authenticator Types by Risk Tier
- Privileged and high-value roles such as domain administrators, cloud administrators, finance approvers and executives: device-bound passkeys on hardware security keys or managed devices, with at least two registered authenticators each.
- General workforce: platform passkeys on managed laptops and phones, synced within an enterprise-controlled credential manager where policy allows.
- Frontline and shared-device workers: hardware keys or badge-based FIDO authenticators, since personal phones may not be appropriate.
- Contractors and partners: require phishing-resistant methods through federation or issued keys rather than exempting them.
3. Pilot, Then Make Passkeys the Default
Start with IT and security staff, then a business unit that is high-risk but supportive, such as finance. Measure registration completion, sign-in success, help-desk tickets and time to sign in. Then flip the default so that new enrollments create passkeys first, and use conditional access policies to require phishing-resistant authentication for sensitive applications and administrative actions.
4. Retire Phishable Fallbacks
The security benefit only arrives when attackers can no longer downgrade to SMS, codes or push. Set dates to remove phishable methods for each user group, beginning with privileged accounts. Monitor for sign-ins that still use legacy factors and treat them as exceptions requiring justification.
Close the Recovery and Help-Desk Gap
As authentication hardens, attackers move to the softest remaining path: convincing a human to reset access. An organization with passkeys everywhere but a help desk that will register a new device after a confident phone call has simply relocated the vulnerability. AI voice cloning makes this path more dangerous, because an attacker can sound like the employee or like their manager.
- Register multiple authenticators at enrollment so that most lost-device cases are self-service through a second passkey rather than a help-desk reset.
- Use strong identity verification for resets: document verification with liveness checks, in-person verification for high-risk roles, or verification by the user’s manager through an already-authenticated channel.
- Never rely on voice or video alone as proof of identity, and never accept caller-supplied phone numbers or meeting links for callbacks.
- Issue time-limited, single-use recovery credentials that can only be used to register a new passkey, and notify the user on all registered channels when a reset occurs.
- Alert on authenticator changes: new MFA registrations, especially followed by unusual sign-ins or new federation settings, are a classic indicator of takeover.
Beyond Login: Defending Decisions Against Deepfakes
Passkeys stop credential theft, but the Arup case shows that some deepfake attacks never touch a login page. They target payments, data releases and urgent requests from apparent executives. The answer is to bind high-risk decisions to authenticated workflows rather than to the apparent identity of a voice or face.
- Require that payment changes, new vendor bank details and large transfers are initiated and approved inside systems where every approver signs in with phishing-resistant authentication, not over email, chat or calls.
- Adopt out-of-band verification using contact details from a trusted directory, never those provided in the request.
- Use pre-agreed verification phrases or challenge questions for executive requests made by voice, and empower staff to pause any urgent request without penalty.
- Update awareness training to show real examples of cloned voices and synthetic video, and shift the message from spotting the fake to following the process.
AI makes impersonation cheap, so security can no longer depend on people recognizing a fake. Passkeys remove the secret attackers want, and authenticated workflows remove the decisions they try to hijack.
A Worked Example: A Distributed Professional Services Firm
Imagine a hypothetical professional services firm with 3,000 employees across North America, Europe and Asia-Pacific, using a cloud identity provider for SSO, a handful of legacy on-premises applications and a centralized service desk. It already uses push-based MFA.
In month one the identity team inventories applications and finds that about four in five are federated through SSO and can enforce passkeys immediately. The remainder include an older HR system and a VPN that need upgrades. Administrators and finance staff receive two hardware security keys each; everyone else is invited to register a platform passkey on their managed laptop and phone.
After a six-week pilot with IT and finance, the firm makes passkeys the default for new enrollments and introduces a conditional access policy requiring phishing-resistant authentication for email, finance systems and admin consoles. The service desk adopts a new reset procedure: identity document verification with liveness for standard staff, manager confirmation through an authenticated workflow for privileged users, and a single-use recovery link that can only enroll a new passkey. The finance team moves vendor bank-detail changes into the ERP with dual approval, and any request arriving by call or chat must be re-initiated there.
Within two quarters, push MFA is removed for privileged accounts and scheduled for removal for the rest of the workforce. Security operations add detections for new authenticator registrations and unusual recovery events. When a staff member later receives a convincing voicemail from a cloned executive voice asking for an urgent transfer, the process itself blocks it: there is no path to move money outside the authenticated workflow.
Pitfalls to Avoid
- Leaving phishable fallbacks in place indefinitely, which lets attackers simply choose the weaker path.
- Ignoring recovery design, turning the help desk into the new primary attack surface.
- Forgetting non-browser access such as VPN clients, legacy thick clients, command-line tools and API tokens, which can bypass the passkey policy entirely.
- Allowing unmanaged personal credential managers for privileged accounts without assessing sync and export controls.
- Treating session tokens as safe. Pair passkeys with token protection, short session lifetimes and device compliance checks, because stolen cookies from infected endpoints remain a risk.
Conclusion
Generative AI has tilted social engineering in the attacker’s favor, and better training alone cannot close that gap. Phishing-resistant authentication changes the equation by making the most common goal of phishing, stealing a reusable credential, technically impossible. In 2026 the standards, platforms and identity providers are ready, and adoption data shows the remaining obstacles are legacy integration, budget and recovery design rather than user reluctance. Organizations that pair passkeys with hardened recovery procedures and authenticated approval workflows for high-risk decisions will be resilient not only to today’s AI-enabled phishing, but to the next generation of synthetic impersonation.
How Hanoi Technology Can Help
Hanoi Technology helps organizations plan and deliver passkey and phishing-resistant MFA rollouts, secure recovery processes and identity monitoring through projects, dedicated teams or managed services.