Security

Passkeys, MFA and the phishing-resistant question

Passkeys, MFA and the phishing-resistant question

7 min read·1509 words·Updated 25 September 2026

The one-time code your bank sends by text was the best available defence in 2012 and it is, today, the thing I watch attackers walk through most often. Not by breaking it. By asking for it. A convincing login page, a victim who types their password and then the six digits from the text, and a script on the other end that relays both to the real site within the 30-second window. That’s it. That’s the attack, it’s automated, it’s sold as a service, and it works against the majority of multi-factor authentication deployed in Australian organisations right now.

The fix has a name, phishing-resistant MFA, and it’s been in the Australian Signals Directorate’s Essential Eight at the higher maturity levels since the 2023 revision. What follows is what the term actually means, why it matters more than the vendor category it’s usually sold in, and how to deploy it without the finance team ringing the helpdesk in tears.

What “phishing-resistant” means, precisely #

An authentication factor is phishing-resistant if it can’t be captured and replayed by a fake site, even with the user’s full cooperation. That rules out anything the user reads and types: SMS codes, authenticator app codes, codes from a token with a display. It rules out most push notifications, because a user under pressure will approve a prompt they didn’t initiate, and attackers have learned to send fifty of them until someone does. What’s left is authentication that’s cryptographically bound to the real site: the browser proves to the server that it holds a key, and the key only works for the genuine domain, so a look-alike site gets nothing useful.

In practice that means FIDO2 credentials, either on a hardware key (the small USB or NFC device) or as a passkey stored on a phone, a laptop or a password manager, plus the older smart-card and certificate-based systems that government and defence have used for years. Passkeys are the same underlying standard, WebAuthn, wrapped in a consumer-friendly sync model, and they’re the reason this is finally practical for organisations that couldn’t hand out hardware keys to everyone.

Keys Padlocks — phishing-resistant MFA
Photo: Leeroy via StockSnap (CC0 1.0)

What the Essential Eight expects #

The Essential Eight is ASD’s baseline of eight mitigation strategies with three maturity levels each, and multi-factor authentication is one of the eight. At maturity level one, MFA is required for remote access and for privileged users, and most methods count. By maturity level two, the guidance expects phishing-resistant MFA for privileged access and for users of important data repositories, and maturity level three extends it further. The maturity levels aren’t law for the private sector, but they’re the yardstick that insurers, government procurement and, increasingly, the boards of regulated entities use to ask whether an organisation is doing the basics. APRA’s prudential standard on information security, CPS 234, doesn’t name a technology; it asks whether the controls are commensurate with the threat, and the threat is the relay attack in my first paragraph.

The gap I see most often is an organisation that reports “MFA everywhere” to its board and means push notifications everywhere. That’s a real improvement over passwords alone. It’s not what the higher maturity levels mean and it’s not what stops the current attacks, and the board deserves to know the difference.

Where to start: the accounts that matter #

Don’t start with everyone. Start with the accounts whose compromise ends the company: domain and cloud tenant administrators, the identity provider’s own admin accounts, the people who can move money, the backup system’s credentials, and the break-glass accounts that exist for emergencies. That’s usually fewer than a hundred people in a mid-sized organisation. Give them hardware keys, two each so a lost one isn’t an outage, enforce them at the identity provider, and remove every other method from those accounts, because an account that accepts a hardware key or an SMS code is protected by the SMS code.

The hardest conversation is usually with the executives, who want the same access from a phone on a beach that they have at a desk. A passkey on the phone solves this without weakening anything, and the lost-phone scenario is handled by the second credential. I’ve never had an executive object once they’ve used it, and I’ve had several ask why it took so long.

Why codes can be relayed and passkeys can't. The same login, two kinds of second factor
Diagram: Technology Spectator

Rolling it out to everyone else #

For the general population, passkeys through the platform they already use, a managed laptop with a fingerprint reader or a phone with biometrics, are the practical route. The identity provider does the work; the user sees “sign in with your fingerprint” and mostly stops ringing the helpdesk about MFA at all, which is a support saving worth putting in the business case.

Three things go wrong. Shared devices in warehouses, clinics and retail, where a personal biometric doesn’t fit; those need a hardware key on a lanyard or a workable exception. Legacy applications that don’t speak modern authentication and can only be fronted by the identity provider with some engineering; make the list before you start, because that list is the project. And recovery. The user who loses every credential at once needs a way back in that’s secure enough not to be the new phishing target, which means a verified in-person or video process for the high-value accounts and a slower, deliberate one for everyone else. Attackers now target the helpdesk recovery process specifically. Write it down, rehearse it, and don’t let it be “ring Dave”.

The Optus and Medibank lesson, read narrowly #

The two breaches that reset the Australian conversation about all this were the Optus incident in September 2022 and the Medibank incident a month later, both of which exposed the personal information of millions of people and both of which led, directly, to the penalty increases in the Privacy Act at the end of that year and to the broader reform programme since. I won’t relitigate either; the public findings and the court proceedings speak for themselves and the OAIC has published its own actions. The narrow lesson relevant here is that in the Medibank case, according to the regulator’s own published statement of the matter, access was gained using credentials for an account that did not have multi-factor authentication enforced. One account. That is the whole argument for starting with the privileged and remote-access accounts and not stopping until every one of them is covered.

Costing it honestly #

Hardware keys are cheap per unit and expensive in logistics: buying, enrolling, replacing and revoking them for a distributed workforce is a real programme. Passkeys through managed devices are close to free in licensing and cost engineering time in the identity platform and in the legacy applications that have to be fronted. In the business case I’d put the helpdesk saving first, since MFA resets are usually in the top three ticket categories and passkeys reduce them sharply; the insurance premium second, since underwriters now ask the phishing-resistant question directly; and the incident avoided third, because boards are tired of hearing about incidents avoided. It’s still the biggest number.

One implementation detail that bites: conditional access. Enforcing phishing-resistant methods only for “risky” sign-ins, as several identity platforms encourage, leaves the ordinary sign-in from a known device on the weaker method, and the relay attack looks like an ordinary sign-in from a known device, because the attacker’s proxy is passing through the victim’s real browser fingerprint. Enforce the method, not the risk score. The risk score is a useful second signal and a terrible first one.

And keep a register of exceptions with an expiry date on every line. The exception granted to a legacy application in 2024 “until we replace it” is still there in 2027 in every environment I’ve audited, and it’s the door.

What it doesn’t fix #

Phishing-resistant MFA protects the login. It doesn’t protect the session after the login, and the newer attacks steal the session token from the browser instead of the credential, either through malware on the endpoint or a malicious browser extension. Short session lifetimes, device-bound tokens where the provider supports them, and endpoint controls (which are, not coincidentally, several of the other seven strategies in the Essential Eight) close that gap. It also doesn’t protect against the employee who’s persuaded to make a payment by a convincing email from the “CEO”, because no authentication was involved. That’s a process control and a culture, not a key.

The opinion I’ll defend is that the industry has spent a decade selling MFA as a checkbox and it’s now paying for that in incidents. The question to ask a vendor, a colleague or a board paper isn’t “do we have MFA”. It’s “can it be relayed”. If nobody in the room knows, that’s the answer.

My sourdough starter, which is nine years old and has survived three house moves and one incident that kept me away from it for a week, uses a process I’d describe as phishing-resistant: it only responds to flour. There’s a lesson in that somewhere and I’m to tired to find it.

— Tom Abernathy, Security Editor, Technology Spectator

Back to top button