AI Scams Exploit Human Trust as US Fraud Losses Reach $21B

Security researchers at Dark Reading report that artificial intelligence is exceptionally effective at manipulating people, with social engineering and scams rising alongside advances in frontier models. In an interview at Black Hat USA 2026, Fred Heiding of Menlo Park Intelligence, alongside former US National Cyber Director Chris Inglis, described AI-driven deception as an asymmetrical threat: attackers can rely on AI to persuade, while defenders struggle to patch “old mental heuristics” in human targets.

The FBI’s IC3 data underline the cost, with reported fraud to US citizens rising from about $4 billion in 2020 to roughly $21 billion in 2025, much of it tied to AI-enabled schemes.

The article highlights how AI can amplify a scam’s effectiveness across the traditional kill chain—target identification, rapport building, and money exfiltration—by enabling highly believable voice chats, long multi‑turn interactions, and even romance-style manipulation. Evidence includes a Reuters investigation into AI scams targeting seniors, including scammers in Myanmar and anecdotal feedback from senior centres about attackers who know personal details and routines.

Practical responses discussed include leveraging AI defensively (for scam detection and analysis), and policymakers considering measures to restrict emotional attachment to AI, or to require clearer disclosure that a chat is with software rather than a human. The piece stresses that, while blue‑team tools can help, there is a limit to curbing manipulation, and the scale of fraud necessitates a combination of detection, regulation, and public awareness.