Down the Rabbit Hole: Red Team Deep Fakes and the Evolution of Social Engineering

Alethe Denis, a senior security consultant on the red team at Bishop Fox, opens with a story about a real-time voice deep fake used to impersonate a chief executive and ask an employee for financial reports. She then explains the main kinds of synthetic media used in social engineering, from GAN-based video and face swapping to text-to-speech and live voice conversion, and the open-source tooling behind them. She is candid about current limits: real-time inference needs substantial GPU capacity, the delay between a spoken line and its synthetic output can expose a pretext, and video artefacts and newer platform detection features still get in the way.

The middle of the talk looks at how attackers actually use this technology. Opinion manipulation and monetised fraud dominate observed misuse, while classic intrusion sits near the bottom. She walks through the 2024 Hong Kong case, in which a finance worker was persuaded to transfer about $25 million during a video call filled with deepfaked executives. She argues that red-team scenarios often drift toward hype, and that narrow scopes and strict rules of engagement can produce a false sense of security.

Her guidance favours building scenarios from real attacker tactics and business risk, pairing a traditional red team with a deep-fake tabletop exercise, and relying on technical controls such as separation of duties and browsing restrictions alongside training. The closing questions cover verification habits, asking questions that only insiders could answer, and whether voice authentication can be trusted.