If your phone rang at 3 a.m. and a voice that sounded exactly like your daughter was screaming that she’d been kidnapped, would you wire the ransom before you called her back? That is the precise scenario the new consumer app Savi was built to defuse. It launched on July 7, 2026, for iPhone and Android, and its pitch is direct: stop AI-cloned kidnapping and grandparent scams before they reach a bank wire.
The app, and what it actually does
Savi was built by brothers Patrick and Ryan Coughlin, who announced a $7 million seed round led by Acrew Capital with Magnify Ventures, TTCER, and Resolute Ventures participating. Patrick’s prior work was in national cyber defense, Splunk, and as Cisco’s SVP of security products; Ryan worked on consumer products at Apple and Spotify. The trigger for the company was an attack that hit their own family. Their mother received an AI-generated voice call that spoofed his sister’s number, mimicked her voice, and named a local Walmart as the supposed drop-off point. She kept her wits, hung up, and verified the daughter was safe. The brothers decided to build the product they wished had existed during that call.
The app has three layers. It screens texts, voicemails, and incoming calls for known scam patterns before they reach the user. On live calls, a user can add Savi’s agent as a third listener, and the agent flags behavioral tells that show up when an imposter is improvising. The model layer is built around Google’s Gemini, run through an AI gateway that lets the company swap in other voice-detection models. Pricing is $8 per month or $63 per year, and a single plan covers an entire family with no per-seat cap.
The training data comes from a free sister site, Scamwise, which the Coughlins launched roughly four months before the app. As of the Savi launch, Scamwise had received about 100,000 user submissions and was adding roughly 10,000 per week. That corpus is what feeds the detection model. Savi Security, Inc. is also a Better Business Bureau-accredited business, per Scamwise’s footer.
The threat this is built against
The market for the product is not theoretical. McAfee’s Artificial Imposters blog post, summarizing McAfee Labs’ 2023 global study of 7,000 people, found that one in four respondents “had experienced an AI voice cloning scam or knew someone who had.” One in ten said they had personally received a clone, and “77% of those victims said they lost money as a result.” McAfee Labs researchers separately tested voice cloning tools and found that “just three seconds of audio was enough to produce a clone with an 85% voice match to the original.” Three seconds is roughly the length of a single voicemail greeting or a short TikTok clip with someone’s voice in the background. That is the technical reason a “kidnap your daughter at 3 a.m.” call is suddenly feasible. The bar to entry for the attacker is a piece of public audio. The bar to fooling a panicked parent used to be much higher - the data lines up with our earlier look at the AI voice-scam epidemic.
The financial gravity behind these scams is also moving up. The FTC’s March 2025 release of 2024 Consumer Sentinel data reported that consumers filed 2.6 million fraud reports in 2024, with imposter scams the most commonly reported category, at $2.95 billion in reported losses for the year. Government imposter scams alone reached $789 million in losses in 2024, an increase of $171 million over 2023. The FTC identified email as the most-reported contact method for fraud overall in 2024, with phone calls second and text messages third, though the press release notes that victims who interacted with scammers on the phone lost more money per person (a median of $1,500) than any other contact channel. That is the demand Savi is targeting. The pattern the founders describe - a synthetic voice, an urgent story about a kidnapped relative, a money demand before the victim can think - is exactly the high-pressure, real-time scenario the FTC’s imposter-scam category tracks.
Why this is also a privacy story
The product sits on a quiet privacy line. To flag a live call as a scam, Savi has to listen to the call. To screen texts and voicemails, it has to read them. To train and improve its detection, it ingests the submissions Scamwise users hand over. That is a meaningful expansion of the data any individual consumer is sharing with a venture-funded AI startup, and it is also why the Better Business Bureau accreditation, the public corporate entity (Savi Security, Inc.), and the named founders matter: they are the first three things a careful user should be able to verify before adding an AI listener to a private phone call.
The model choice is also worth noting. Savi is not running everything on-device. The article describes Gemini accessed through an AI gateway, which means audio and text payloads for screening are flowing through Google’s infrastructure. That is normal for the category, but it is a real data flow that any current or prospective subscriber should understand. Anyone who handles sensitive voice content as part of their job - lawyers, journalists, doctors, abuse counselors - should read the privacy policy carefully before installing. The same is true for any consumer who is already worried enough about voice cloning to want this app. The defense against a voice clone should not quietly create a new permanent audio pipeline to a third party.
There is also a market question. Once the big platforms ship similar features, the standalone consumer app has a narrower moat. Both Apple and Google already have on-device scam-detection work in progress for iOS and Android. The window where Savi is the only consumer app designed for this threat is short, and the company has telegraphed that by pricing aggressively for the family-plan angle and by owning the Scamwise training corpus.
What This Means
For readers, the practical question is whether Savi is the right tool today or whether the same protection will land inside the phone they already own. The current state is that Savi ships a polished, family-priced product now, and the platform-level equivalents are still arriving. The right decision depends on who in your household is most likely to be targeted, how much audio of their voice is already public, and how comfortable you are with one more AI service listening to your calls. If your household includes an older relative who takes urgent calls from family members at face value, this is one of the few consumer products built specifically for that exposure. The Better Business Bureau accreditation and the named founders are not a guarantee, but they are an above-average level of accountability for the category.
The deeper point is that AI-cloned kidnapping scams are now cheap enough to industrialize - the same arc we tracked in February when deepfake fraud crossed into $25 million single-call territory. Three seconds of audio is enough to clone a voice. A free cloud text-to-speech API is enough to script the call. The only thing standing between the attacker and the bank wire is the few seconds it takes the victim to hang up, verify the relative is safe, and refuse to send money. Savi is betting that a real-time listener trained on 100,000 scam submissions can flag the call during those few seconds. It is a market that did not exist three years ago, and it is going to be one of the first places where a consumer AI product has to prove it can be trusted with the very audio it is trying to protect.
The Bottom Line
Savi is a real product, shipping today, with named founders, $7 million in seed funding, a $63-per-year family plan, and a 100,000-submission scam corpus. The threat it is built against is real, growing, and priced at $2.95 billion a year in reported US losses. Try the free Scamwise site before you commit, and read the privacy policy if anyone in your family handles sensitive calls for a living.