SentiLink CEO: Most Anti-Fraud Applications Are Not What They Claim to Be, Deepfake Threat Underestimated
At the recent Money20/20 fintech conference in Las Vegas, SentiLink CEO Naftali Harris stated bluntly that many anti-fraud software solutions fall far short of expectations in practice, making it difficult to withstand the impact of new types of scams such as deepfakes. He warned that over the next 20 to 30 years, fraud and account takeover issues will become increasingly serious, and even with the deployment of AI defenses, the outlook remains bleak. Oscilar CEO Neha Narkhede shared cases where deepfakes have evolved from one-time impersonations into persistent synthetic identities, and emphasized that human-machine collaboration will be key to combating fraud.
Quick Overview
- Naftali Harris, CEO of SentiLink, said that many current software anti-fraud measures struggle to cope with the growing threat of deepfakes and other scams.
- During a panel discussion at the Money20/20 fintech conference in Las Vegas, Harris said, "A not-so-glorious secret is that not all anti-fraud controls are that effective." He believes that even with AI defenses deployed, the future of anti-fraud is not particularly optimistic.
- Harris predicts that over the next 20 to 30 years, "fraud and account takeovers will become much more severe." He said, "Many of us will fall for these scams because they will be extremely deceptive. This is a world we haven't yet faced."
Deep Insights
According to Keepnet Labs, a developer of anti-phishing platforms, among the many types of scams, deepfake phishing attacks using AI-generated voice and video have surged and are increasingly difficult to identify. Keepnet data shows that the number of deepfake files skyrocketed from 500,000 in 2023 to 8 million last year, with the average loss per deepfake-related incident for businesses approaching $500,000. Additionally, the number of deepfake incidents detected in 2023 was 10 times that of 2022.
Neha Narkhede, CEO of Oscilar, said during the panel discussion, "Deepfakes are still in an emerging stage—not yet rampant." She predicts these attacks will "become mainstream." She added, "What we're seeing in practice is early but clear signs of deepfakes evolving from one-time impersonations to persistent synthetic identities."
During the discussion, Narkhede revealed that this year one of Oscilar's cryptocurrency clients noticed seemingly ideal user growth: hundreds of accounts started with small transactions and then gradually increased withdrawal amounts. "In isolation, everything seemed normal, but when these signals converged on a unified platform, the situation was completely different," she said. "These accounts had similar GPU fingerprints, nearly identical TLS signatures, and matching video encoding patterns. These all indicate they were likely created by the same automated stack." Furthermore, the pattern and speed of transactions were coordinated rather than "organic." "It turned out to be a typical synthetic network driven by deepfakes," she concluded.
Harris pointed out that deepfake fraudsters and other criminals do not lack exploitable personal data, which originates from cybersecurity breaches. "There have been so many data breaches in the past that I actually think additional identity information leaks won't give fraudsters more of an advantage than they already have," he said.
Narkhede predicts that effectively combating fraud will rely on human-AI collaboration. She said that humans "will guide AI agents—AI copilots will handle the heavy lifting while keeping humans in the decision loop." She further noted, "When AI agents start making transactions on our behalf, I think the concept of identity itself will shift from 'who you are' to 'what is transacting on your behalf.'"