← All finance signals
Trust signal

Beyond the Blink Test

A face or fingerprint approves a payment, not just a phone unlock, while a WEF study finds face swaps can fool certain biometric checks under specific conditions.

What is it?

A face or fingerprint now unlocks a payment, not just a phone, but the checks meant to prove the human behind it is real are already beaten by adversarial AI. Platforms answer by moving proof off software, onto the device. Whether liveness can still be proven stays open.

Why now?

Biometrics were adopted for convenience, not built to survive a generative attacker; commercial detection now loses roughly half its lab accuracy once deployed against real deepfakes. The old worry — someone slipping past a check — has inverted into an attacker whose fake passes convincingly. That inversion is forcing the rebuild now.

Who’s building it?

Platform owners rebuilding the credential layer; enterprises complying on a forced deadline set above them.

Platform ownersPasskey vendorsEnterprise ITDeepfake detectorsRegulators
The full read
Why now, who is building it, where the market goes, and what stays unsettled.
Get weekly →
01 · The market
Who’s in it
Builders — device-credential vendors, and separately, biometric-database and liveness-detection firms.
Buyers — enterprises under compliance deadlines, and platforms rebuilding login and payment flows.
Gatekeepers — Microsoft and the other identity platforms setting the enforced migration timeline.
What’s at stake
The split — device-bound credentials carry limited exposure; server-side biometric stores carry far more — decides who owns which half.
Whether liveness detection holds up at the moment of capture.
People whose biometrics read poorly face new payment friction.
What’s unsettled
Whether liveness checks can outpace adversarial AI, or lose the race permanently.
How much legal exposure server-side biometric storage is really worth carrying.
Whether device-bound credentials become the default, leaving biometric databases a shrinking, riskier niche.
02 · How the demand matureda progression, not a timeline
Mandate
A platform sets a forced deadline, moving enterprises off software checks by decree.
Enrolment
Users are auto-enrolled into passkeys as legacy SMS and voice logins are retired.
Liability split
Device-bound credentials and biometric stores diverge on legal exposure, steering where firms build.
Presumption flip
A face or voice moves from trusted by default to unverified until proven live.
03 · Where the market goesthe forks, not a prediction
→Device-bound or server-side — whichever carries less legal exposure over time likely absorbs the market, leaving the other a shrinking, riskier specialism.
→Liveness detection could close the gap on adversarial AI, or lose ground and get abandoned.
→Regulation could formalise the body as a financial credential, with real stakes for those it misreads.
→Platforms, not standards bodies, may end up writing the real spec by enforced deadline.
The open question is whether liveness itself can still be proven against a generative attacker — every other split downstream assumes it can.
04 · The Body as Interface deskadjacent markets, moving now
The Interior Turn
Emotion-reading AI now faces workplace limits.
EU banned workplace emotion-recognition AI from Feb 2025.
European Commission
Ambient Health
Wearables now carry real regulator medical clearance.
Apple Watch, Garmin gained FDA clearance for AFib, sleep apnea.
Pulmonology Advisor
The Verification Ritual
Human-verification badges spread across dating and video apps.
World announced human-check deals with Tinder, Zoom and Docusign.
TechCrunch
05 · The Wire
01
WEF study: face swaps plus camera injection can fool some biometric checks
Infosecurity Magazine ↗
02
Microsoft makes passkeys the Entra ID default from 1 September 2026
Computerworld ↗
03
Visa Payment Passkey is live in India with IDFC First Bank
BusinessToday ↗
04
Open-source deepfake detectors lose about half their AUC on real-world fakes
arXiv (preprint) ↗
The panel on this signal
They agreed on
–The threat inverted: the risk is no longer failing a check, it’s a fake passing one convincingly.
–The split is a liability question, not a technical one.
–Platforms, not regulators, are the ones setting the real migration timeline in practice.
The security researcher disagreed on
–The accuracy drop documented moving from lab to real-world conditions was measured against login authentication; payment-authorisation fraud carries materially higher stakes and attacker incentive, and the same error rate has not been separately validated in that context.
Know more about the method →
✦How this signal was built
01 · Panel
A Delphi panel of 4 AI analysts, each reasoning independently.
→
02 · Debate
They challenge and revise each other’s reads over two rounds.
→
03 · Evidence
Every claim is tied to a cited, linked source on The Wire.
→
04 · Synthesis
A moderator writes it into one read, disagreement and all.