When Patrick Coughlin’s mother called him two years ago, she was in a panic. A man had just told her he had kidnapped her daughter. She heard what sounded like her daughter’s voice screaming, then pleading, then the man demanded $1,200 or he would kill her in a Walmart parking lot. The caller ID showed her daughter’s number. The voice sounded exactly right. The Walmart reference was accurate. None of it was real.
The kidnapping was an AI-generated scam. Coughlin’s mother kept her head, called her daughter directly, and confirmed she was safe. But the incident stuck with Coughlin, who at the time was senior vice president of security products at Cisco. He had spent his career watching sophisticated cyberattacks hit governments and Fortune 500 companies. Now the same tools were being aimed at his mom.
That experience led Coughlin and his brother Ryan to build Savi Security. As reported by TechCrunch, the company has raised $7 million in seed funding led by Acrew Capital, with Magnify Ventures, TTCER, and Resolute Ventures also participating. The app is now live on iOS and Android.
The brothers bring serious credentials to the problem. Patrick built cloud security startup TruSTAR, which Splunk acquired for a reported $82 million in 2021, before Cisco bought Splunk in 2024. Ryan spent years on consumer products at Apple and Spotify. Together, they’re applying enterprise-grade security thinking to a problem that is hitting ordinary people at scale.
The scale of that problem is growing fast. The FTC reported that people lost $3.5 billion to imposter scams in 2025, three times the figure from 2020. The rise tracks almost exactly with the wider availability of cheap generative AI tools. Before AI, running a convincing voice-clone scam required significant resources and technical skill, which meant fraudsters mostly targeted companies or government agencies where the payoff justified the effort. That calculation has completely flipped.
“You can clone a voice off three seconds of audio, off a publicly available social media post,” Coughlin said. The research material is everywhere: videos of parents at school sports events, casual social media clips, voice notes. Scammers can now build a convincing fake from whatever someone has already shared publicly, and it costs them almost nothing.
The vulnerability is not limited to older generations either. Research from Malwarebytes published in 2025 found that Gen Z was targeted by text scams more than any other age group, and fell for them roughly 25% of the time. This is not a problem with a simple demographic profile.
Before building the paid app, the Coughlin brothers tested their scam-detection model by launching a free, anonymous website called Scamwise. No registration required. Users could upload suspicious texts, images, or emails and get an assessment of whether the content was likely fraudulent. The site attracted 50,000 submissions within four months and is now adding around 10,000 per week, with total submissions reaching 100,000 by launch day. That volume gave Savi a steady supply of real-world data to train its detection model.
The app itself can screen three types of content:
- Incoming text messages
- Voicemails
- Live phone calls
Text and voicemail screening is available in several other security products already, but live-call monitoring is where Savi does something different. If a user feels uncertain during a phone call, they can add Savi’s AI agent as a silent listener. The system monitors the conversation in real time, watching for behavioral patterns that suggest a scam is in progress, and alerts the user while the call is still happening.
Savi currently relies primarily on Google’s Gemini but has built its system on an AI gateway architecture that lets it swap in other models as needed, including tools built specifically for voice detection.
Pricing is also structured differently from most security software. The $8 per month plan (or $63 per year) covers an entire household with no cap on the number of users. One subscription can protect parents, children, grandparents, and anyone else the account holder wants to add. That design is deliberate. The people most at risk from this kind of scam are often older family members who may not be managing their own security software, but who someone in the family wants to protect.
Coughlin frames the broader problem in direct terms. AI is not just helping existing criminal networks run more efficient scams. It is pulling in new people who would not previously have attempted fraud, because the technical barrier to deceiving someone has dropped so far. The organized syndicates are getting more powerful, and the pool of people attempting fraud is getting larger at the same time.
Savi’s bet is that the same AI driving that threat can also be used to catch it in real time, the same way antivirus software evolved to keep pace with malware. The difference is that this time, the target is a phone call happening right now, not a file that already landed on a hard drive.