Lessons from building AI vishing simulations: help desk resets are the soft spot. What have your vishing tests turned up?
retained text
Disclosure: I work on the security team at a security vendor (Fable). No link or pitch here; I’m curious about practitioner experience. As email filtering gets better, more social engineering is moving to the phone. The pattern we keep seeing is someone calling the help desk, saying they lost their phone, and asking for an MFA or password reset. AI voice makes it cheap to try that at scale, so attackers just keep calling until they reach someone who’s undertrained. A few things that surprised us while building realistic vishing simulations: * **The first few seconds matter most.** Lag at the very start of a call is what gives a fake away. Once the conversation gets going, people tolerate a lot more. * **Realism depends on org context.** If the caller’s name isn’t in the company directory, the help desk catches it right away, because that’s the first thing they check. Generic scenarios teach the wrong lesson. * **Knowing when to stop is the hard part.** Unlike phishing tests, orgs disagree a lot about how far a vishing test should go before it breaks character and switches to coaching. There’s no consensus yet. For those who’ve run vishing exercises: what did you find that email phishing tests missed? And if you decided against them, how do you assess phone-based social engineering risk instead (help desk verification audits, red team, something else)?
document fetched 2026-10-30 00:23 · canonical https://reddit.com/r/cybersecurity/comments/1xily3g/lessons_from_building_ai_vishing_simulations_help