What deepfake security awareness simulations look like in 2025 (www.itbrew.com)

Security awareness vendors are already turning deepfakes into immersive training tools to prepare employees for increasingly realistic AI-driven scams. Reporters who trialed demos from Adaptive Security, Breacher.ai and Arsen Security saw convincing video clones, Teams vishing calls and phone-based “IT” calls generated from as little as a single photo and ~10 seconds of audio (Adaptive says a demo can take about five minutes to produce). These vendor simulations replicate real attacker tactics—urgency, social pressure and plausible context—and mirror real-world impact: Resemble.ai reports $347.2M in direct deepfake losses in Q2 2025. Vendors also report high initial failure rates for simulated attacks (a “high double-digit” percentage) that can fall below 10% with repeated, interactive training. The demos highlight two urgent technical and operational implications: humans are rapidly becoming unreliable detectors (iProov’s CTO demonstrated face-swapping that even experts struggle to spot; an on-site quiz yielded only 7/10), and attackers can use agentic voice models to answer questions and defuse skepticism. Practical defenses emphasized by vendors include out-of-band verification, throttling urgent requests, red-team simulations, and repeated experiential training—plus technical measures such as stronger MFA, anomaly detection and identity verification. As deepfakes approach perceptual parity, the industry may shift toward more in-person verification and robust protocol-driven checks rather than relying on visual or auditory intuition alone.