What changed
Amazon Polly launched gated voice cloning on its generative engine in seven languages. A few minutes of recordings from a consenting speaker create an encrypted, account-controlled voice; AWS says the source recordings are deleted after creation and are not used to train Polly models.
Why it matters
Production voice cloning becomes an API-accessible enterprise capability, so consent evidence, speaker authorization, key access, output disclosure and revocation must be part of the deployment rather than an afterthought.
What to do
Request access only with a documented speaker-consent and revocation process, isolate cloned-voice permissions, watermark or disclose synthetic output where appropriate, and test deletion, misuse escalation and account-compromise recovery.
Caveat
The release is generally available but gated to approved accounts and named Regions. Naturalness, impersonation resistance and abuse controls are AWS claims without independent results in this record.