Passkeys for a world where the phishing writes itself
Why passkeys defeat phishing at machine scale, and the honest list of what they do not fix: stolen sessions, infostealers, and agents holding your logins.
By Priya Ramaswamy-Quill, Editor, Actually Useful4 min read
The case for passkeys used to be made against a phishing email with a typo in it. That email is gone. The one replacing it is written by a model, in your colleague's register, referencing the project you are actually on, and if you hesitate, a voice that sounds like your manager calls to walk you through it. Training people to spot the fake was always a losing strategy. It is now a losing strategy against an opponent that does not get tired.
A passkey is a credential that does not care how convincing the attacker is.
The mechanism
When you create a passkey for a site, your device generates a pair of cryptographic keys: a private key that stays on your device or in your synced keychain, and a public key the site stores with your account. At login, the site sends a random challenge; your device signs it with the private key after you unlock with a fingerprint, face, or PIN; the site checks the signature against the public key.
Three properties follow.
- Nothing secret is sent. The site holds only a public key. A breach of the site yields nothing that logs anyone in.
- Each passkey is unique to one site. Nothing is reused, so nothing is replayed elsewhere.
- It is bound to the site's origin. The browser or operating system will only use a passkey on the exact domain it was created for. A lookalike site at a different domain cannot ask for it.
Why that matters now
The third property changes the economics. Model-written phishing is good because it persuades. Origin binding makes persuasion irrelevant: the user can be completely fooled, click the link, want to log in, and the passkey still will not sign for the wrong domain. There is no password to type into the fake page and no one-time code to read to the voice on the phone. The attacker who spent the effort to clone your manager gets, at the end of the call, nothing to relay.
This is also why phishing-resistant MFA is the fix for push bombing and help-desk social engineering. Those attacks work by extracting something from a person. A passkey gives the person nothing to hand over.
Synced versus device-bound
Most passkeys today are synced through a platform keychain or password manager, encrypted end to end across your devices. That is far more usable and recoverable, and it means your passkeys are only as secure as the account that syncs them. Protect that root account with a long unique passphrase, phishing-resistant MFA, and recovery options you control. Device-bound passkeys on hardware keys remain the right choice for the highest-risk accounts: domain admin, cloud root, the identity provider itself.
What passkeys do not fix
The honest list is longer than it was, because the attackers have moved.
- Session token theft. Once you are logged in, your browser holds a session cookie and the passkey has done its job. Anything that steals the cookie impersonates you without touching the credential. Passkeys protect the front door, not the house.
- Infostealers. Malware that harvests browser cookies, saved sessions, and local tokens is now the main way into accounts that cannot be phished. A model-written lure is a good way to get the stealer onto the machine in the first place.
- AI agents holding long-lived sessions. Agents act on your behalf through delegated tokens and persistent browser sessions. Those are exactly the artifacts infostealers want and exactly what a prompt injection can exploit, and a passkey on the front door does nothing once an agent is inside holding a token that lasts for weeks. An agent with your session is a session theft you authorized.
- Account recovery. If a site lets you recover with email or SMS, that path is weaker than the passkey and is where attackers, and their voice agents, will go.
- Fallback passwords. Many sites add a passkey beside the password instead of replacing it. The password is still phishable. Remove it where the site allows.
- Device compromise. A device the attacker controls can approve passkey use. Nothing fixes this except not having compromised devices.
What to actually do
- Enable passkeys on email, your password manager, your bank, and your cloud and social accounts, in that order. Email first: it is the recovery path for everything else.
- For administrative accounts at work, require hardware keys.
- After adding a passkey, review the account's recovery and fallback options and remove the weak ones. Assume the attacker will call the help desk in your voice.
- Shorten session lifetimes for anything that matters, bind sessions to the device where the platform supports it, and treat an agent's delegated token as a credential to be scoped, expired, and revoked, not a convenience.
- Put endpoint protection and browser hardening on the same list as passkeys. The credential is now the strong link; the machine it lives on is the weak one.
- Do not wait for perfection. A passkey plus a leftover password is still strictly better than the password alone, because the passkey is what you use day to day, and the thing you use day to day is the thing the model is writing to.
About the writer
Editor, Actually Useful
Priya runs the section of the magazine that is allowed to be helpful. A former incident responder, she writes the checklists she wishes someone had handed her at 3 a.m.