A convenient factor with a specific, real weakness
Voice authentication verifies identity based on vocal characteristics, which suits interfaces where typing a password or presenting a device isn't practical — a hands-free assistant, a phone-based system, a smart speaker. The convenience comes with a genuine cost: a voice is comparatively easy to capture, and increasingly straightforward to synthesize convincingly, in ways that most other biometric or possession-based factors are considerably more resistant to.
A system that treats a voice match as sufficient proof for a consequential action is trusting a factor that's weaker than it may appear, especially as voice synthesis technology continues to improve. Treating it as adequate for identifying a likely speaker in a low-stakes interaction is a reasonable use; treating it as sufficient authorization for anything genuinely consequential generally isn't.
Matching the interface's constraints to what it's authorizing
The practical response scopes voice-driven interfaces to lower-stakes actions by default, and requires an additional factor — delivered through a different channel entirely — before a voice interface is allowed to authorize anything with real consequences. This mirrors the same underlying idea behind step-up authentication generally: the required strength of proof should track the stakes of the action, and a hands-free interface's inherent limitations are a reason to set that bar higher for it specifically, not lower.