A startup just put real money behind a problem most of us have only worried about in the abstract: AI scams that sound exactly like the people we love. Savi Security launched its consumer app for iPhone and Android on Tuesday and announced a $7 million seed round, according to TechCrunch AI. The round was led by Acrew Capital, with participation from Magnify Ventures, TTCER, and Resolute Ventures.
What stands out here is the origin story, and it’s a gut punch.
The call that started it
About two years ago, co-founder Patrick Coughlin’s mother got a phone call. The caller ID showed his sister’s number. She heard what sounded like his sister’s voice saying “Mom, they’ve got me,” then a scream, then a man demanding $1,200 or he’d kill her daughter in the parking lot of the local Walmart she actually visited, TechCrunch AI reports.
It was fake. The whole thing, generated by AI. His mom kept calm, called the daughter, and found her safe. But the number, the voice, and the location were all spoofed convincingly enough to terrify a level-headed person.
Coughlin isn’t a random founder either. He ran security products at Cisco, sold his cloud security startup TruSTAR to Splunk for a reported $82 million, and worked in national cyber defense. His brother Ryan came from consumer product roles at Apple and Spotify.
Why this matters now
Here’s the shift that makes this urgent. The kind of targeted, research-heavy attacks that used to be aimed only at governments and Fortune 500 companies are now cheap enough to point at your parents.
“You can clone a voice off three seconds of audio, off a publicly available social media post,” Coughlin told TechCrunch AI. Every time someone narrates a kid’s football game on Facebook, they leave a usable voice sample in the open.
The numbers back him up:
- The FTC says people lost $3.5 billion to imposter scams in 2025, triple the 2020 figure.
- Older Americans report the most losses, but Gen Z isn’t safe. Malwarebytes research found Gen Z was targeted with text scams more than other generations and fell for them about 25% of the time.
Before cheap, powerful language models, running these cons on regular people wasn’t worth the effort. That math has flipped.
What Savi actually does
The app screens texts, voicemails, and incoming calls for scams. Plenty of products do that. Savi’s differentiator is live-call monitoring.
During a suspicious call, you can add Savi’s live agent as a silent listener. It watches for behavioral tells that signal a con while the call is still happening, so you get a warning in the moment instead of a regretful lesson afterward.
Under the hood, Savi mostly runs on Google’s Gemini but built on an AI gateway, so it can swap in other models, including voice-detection specialists, as needed.
The company also trained its detection model on real-world data from a free tool it launched four months ago called Scamwise. It’s anonymous, no signup, upload a suspicious text, photo, or email and it tells you if it looks bogus. That tool has hit 100,000 submissions and grows by roughly 10,000 a week, per TechCrunch AI.
The pricing angle
The business model is worth noting because it’s built around how these scams actually spread. Savi charges $8 a month, or $63 a year, to cover an entire family with no cap on users. One plan covers kids, a spouse, parents, and, as Coughlin puts it, “that uncle who always seems to need tech support.”
That’s a smart read. Scammers target the least protected person in a family, so protection that stops at one device misses the point.
What comes next
Savi’s own framing is that it’s building a new kind of antivirus, AI defending people in real time while criminals use AI to swindle them. Expect that framing to spread. As voice cloning gets cheaper and the FTC’s loss numbers keep climbing, real-time scam detection is likely to move from niche product to a category incumbents like Malwarebytes and the phone carriers fight over.
For now, the practical takeaway is simpler. Agree on a family code word, verify before you pay, and assume a familiar voice on the phone can be faked. You can find the full details at the original source.