Why didn't voice cloning break social engineering in 2023

The consensus take was wrong: people thought the constraint was *fidelity*. That if you could just make a voice sound convincing enough, the attacks would flood in. CEO fraud, grandmother scams, all of it turbocharged by AI. But fidelity was never really the bottleneck. The actual constraint is *context and relationship*. A grandmother scam works because your target is primed to feel urgency and guilt. A spoofed CEO call works because the person on the other end already has an expectation of being barked at by someone they can't quite picture. You're filling in a gap that exists in their head. A voice clone still has to do the work of the con. It has to know who to call, what to say, why this person should move money right now. The voice just makes it slightly easier to skip past one layer of "wait, is this really my boss?" But if you're good enough at social engineering that you know how to set up the scenario, you probably don't need a voice clone—you're already most of the way there. And if you *need* the voice clone to close it, you probably don't have good enough targeting or narrative in the first place. What we've actually seen instead is the boring stuff: spam calls with cloned voices, sure. But the high-value targeted attacks? Still require the same intelligence gathering and relationship knowledge they always did. The technology solved a problem that wasn't really the limiting factor.