Fighting AI Scams With AI: That’s This Startup’s Mission [CNET]

Whether it’s a Nigerian prince asking for an advance or a threatening message supposedly from the IRS about back taxes, we’ve all likely been targets of scammers at one time or another.

It’s big business for criminals: Despite heightened awareness, Americans lost $10 billion to scams like these in 2023 alone.

The problem is getting worse for multiple reasons, including a relative newcomer: generative AI. The same technology we use to perfect our resumes, generate realistic images and write funny poetry can be used by scammers to swindle us. With the rise of popular chatbots like OpenAI’s ChatGPT and Google Gemini, scammers have their own superhero sidekicks to compose believable emails and text messages soliciting money.

“We like to say that the bad guys have started using AI and the good guys are using education,” said Scamnetic CEO Al Pascual. “Your bank is telling you, ‘We’ll never do this. We’ll never do that. Never click on this.'” 

While financial services firms use AI internally to identify and prevent scams, we consumers haven’t had a lot of options to use AI to protect ourselves. But now that’s starting to change.

While he was a senior principal and enterprise solutions lead at credit reporting agency TransUnion in 2023, Pascual said he noticed scams were getting worse.

“Something happened with COVID where criminals had this epiphany … especially after the whole thing subsided with government benefits fraud … it seemed like they realized it’s just easier to ask a bunch of people for money because eventually someone’s going to say yes,” he said. “I realized that the opportunity was pretty significant.”

TransUnion isn’t in the scam-fighting business, so Pascual exited and founded Scamnetic in September 2023.

KnowScam and IDeveryone

The startup is focused on two primary areas.

The first is inbound communication like email, text messages and social media posts. Scamnetic analyzes these messages to determine if they’re legitimate or malicious with a product called KnowScam.

Its model uses machine learning and natural language processing to score content based on what it learned in training that included “thousands of good messages and thousands of bad ones,” Pascual said.

(Machine learning is a branch of AI that taps into data and algorithms to mimic the way we humans learn. Natural language processing uses machine learning to help machines understand and communicate using human language.)

KnowScam scores messages on a scale of one, which is low risk, to four, which is high risk. This score is based on message details like sender, images, links and attachments and whether any of these components can be tied to patterns for past scams.

Scamnetic can also help you manage risk when you’re reaching out to people you don’t know, such as on dating sites or e-commerce platforms like Facebook Marketplace, with a product called IDeveryone.

Before you, say, send money to a stranger with a product for sale, you can ask them to verify themselves with this tool. After adding their phone number or email, the person in question receives a message asking for their name, location and permission.

“This is a permission process because we go to non-public data sources, at which point we verify you in the background,” Pascual said.

Tampa, Florida-based Scamnetic plans to launch a direct-to-consumer subscription service in the fourth quarter. For now, it’s working with partners like banks, credit unions, telecoms and insurance companies, which offer KnowScam and IDeveryone to their customers as either a free benefit or as part of a premium package.

Deepfakes

Moving forward, Pascual wants to expand the channels Scamentic covers, including more activity in social media. He also plans to get into video communication to help users detect deepfakes, or images and videos that have been altered to display something that never really happened.

“That’s a big thing on our roadmap for the relatively near term because that’s an impossible problem for consumers,” he said of deepfakes. “It’s one thing to look at an email and say, ‘Does this look like a good PayPal logo?’ But if you see someone on screen in video — and some of the high-quality ones are really good — it can be almost impossible for someone to know.”

Scamnetic raised $1.35 million in a pre-seed round in March.

Investors include early-stage investment firm Ruxton Ventures, Roo Capital and SaaS Ventures, as well as cybersecurity executive Ori Eisen.

“Odds are, we will continue to take funding,” Pascual said.

Other Services & Software