Google Built a Lie Detector for Your Phone. The Lies Are Getting Smarter.

Here's the uncomfortable truth buried inside what looks like a straightforward product update: Google just shipped a feature designed to catch AI, using AI, to protect you from AI. The recursion is dizzying, and nobody in the coverage wants to sit with it long enough to feel what it means.

The feature itself is legitimate. Google's Phone app will now flag calls that appear to come from one of your saved contacts but fail a silent background check — a "confirmation signal," as Wired described it, sent through RCS to verify that whoever's ringing you is actually who the number suggests. If the signal doesn't come back clean, the app raises a flag. The idea is to catch the move that scammers have been running with increasing sophistication: spoofing a trusted number, then layering AI-generated voice on top so the caller sounds like your mother, your boss, your bank.

The FBI reportedly put American losses to AI-assisted scams at over $893 million in 2025, according to The Verge's coverage. That number earns the feature its existence. This isn't a solution chasing a hypothetical problem.

The Signal and the Gap

What nobody is saying loudly enough is what the confirmation signal actually depends on: RCS infrastructure and Android 12 or later, per Wired. That's a real constraint. RCS rollout has been uneven. Older devices don't qualify. And the person most likely to get targeted by a voice-clone scam — an older relative, someone on an aging handset, someone whose carrier still treats RCS like a luxury — may be exactly the person this feature doesn't reach.

The scammer targeting your grandmother probably isn't worried about RCS verification.

TechCrunch noted the behavioral shift already underway: as people increasingly refuse to answer calls from unknown numbers, scammers have pivoted to spoofing numbers people do recognize. This feature addresses that pivot directly, which is credit where it's due. But a pivot met with a countermeasure is just the beginning of a longer game. Deepfake voice technology isn't standing still while Google ships an Android update. The models generating fake voices are trained continuously. Detection features ship on quarterly cycles. The asymmetry is obvious.

What This Is, Really

Ars Technica framed the feature inside Google's broader June Android drop — more scam detection, more AirDrop-style sharing, more AI features across the board. 9to5Google noted Google Photos and Google Play Books are also getting AI updates in the same wave. That context matters, because it tells you something about how Google is positioning this: not as an emergency patch, but as a product feature. Part of a drop. Announced alongside convenience tools.

That's either reassuring — they've got this under control, it's infrastructure now — or it's a tell. Emergency responses don't get bundled with book-reading features. Coordinated product drops suggest confidence. Confidence, in a space where the threat is actively learning, is worth examining.

The silent confirmation signal is clever engineering. The premise — that a legitimate caller will have an RCS-compatible device and a carrier that plays along, and that this combination can be verified fast enough to matter — is sound in controlled conditions. Whether it holds as deepfake technology gets cheaper and more accessible, as carriers lag on infrastructure, as the spoofing techniques evolve to probe exactly the gaps in this system: that's the question the coverage raises without quite asking.

Google built a door lock. The people picking locks just got a new set of tools.

The feature is worth having. Turn it on. Just don't let the flag become the thing you trust instead of your own judgment — because the next version of the scam will probably know how to wave back.