AI voice cloning is the AI replication of a specific person's voice, treated as a distinct harm surface within the broader category of synthetic media and deepfakes. Cloned voices have been used in financial scams, political deepfakes, harassment, and other nonconsensual audio.
Scam and impersonation uses
Voice cloning has been deployed across several impersonation patterns. In so-called "grandparent scams," the cloned voice of a child or grandchild is used to portray a family member in distress. In an enterprise context, cloned executive voices are used to social-engineer finance staff into authorizing unauthorized transfers. Political impersonation has also occurred: the New Hampshire robocall impersonating Joe Biden (January 2024) established a template for cloned-voice political deepfakes.
Individual losses from family-impersonation scams continue to be reported. A Poway, California man lost $18,000 in savings on June 22, 2026 to a phone scam he believes used an AI clone of his daughter's crying voice, paired with knowledge of her childhood medical condition (Source: fox5sandiego.com).
Consumer Reports assessment
Consumer Reports conducted a consumer-protection assessment of AI voice-cloning products and documented consent-verification gaps across most commercial voice-cloning offerings. The assessment found that abuse-prevention guardrails were inconsistent and often trivially bypassed, and that consumer awareness of the risk remained low despite widespread deployment.
Industry landscape
ElevenLabs is described as setting the industry standard for voice-AI quality, offering both enterprise and consumer tiers. OpenAI deploys voice capabilities to consumers through ChatGPT Voice Mode, which is restricted against voice cloning. Other competitors include Resemble.AI, Descript Overdub, and Meta Voicebox. Open-weight voice models are downloadable and bypass commercial safeguards entirely.
Legal and regulatory regime
The TAKE IT DOWN Act (federal, May 2025) criminalizes nonconsensual intimate deepfakes, with text that includes audio. State-level deepfake statutes in Minnesota, Washington, Texas, and California address synthetic media, and many of them cover voice. The proposed federal NO FAKES Act specifically targets voice and likeness replication without consent (see planned NO FAKES Act (federal, proposed)).
Among platform responses, YouTube stated support for both the NO FAKES Act and the TAKE IT DOWN Act on April 9, 2025, and announced an ongoing partnership with CAA on likeness management on December 17, 2024. On September 5, 2024, YouTube introduced Content ID for synthetic content, providing synthetic-content detection and likeness-rights tools.
Consumer-protection enforcement petitions
The Consumer Federation of America, working with students from UCLA Law School's Information Policy Lab, filed a complaint on July 27, 2026 with the FTC and state attorneys general alleging that the voice-cloning platform Speechify violates Section 5(a) of the FTC Act and state UDAP and digital-forgery statutes. The claimed defect is consent architecture: the platform is alleged to allow voice clones from seconds of audio behind little more than a self-attestation checkbox. The complaint cites FTC data showing consumers reported losing $3.5 billion to impersonation scams in 2025, and older Americans $445 million in 2024 (Source: consumerfed.org).
Detection and recognition
A Microsoft finding (July 28, 2025) reported that "most people can't spot AI photos," with similarly limited human capability for detecting AI-cloned audio, particularly in adversarial contexts such as scam calls.
Relationships
- supports: TAKE IT DOWN Act, Synthetic Media / Deepfakes, State Deepfake Statutes (MN, WA, TX, CA), planned NO FAKES Act (federal, proposed).
- related: ElevenLabs, Reality Defender, Data Provenance, C2PA, and Watermarking (provenance + detection), Sora and Veo (Video Generation Models) (video analog), California AI Transparency Act (SB 942) (CA AI Transparency Act watermarking), APA Health Advisories on AI and Adolescent / Mental-Health Well-being (2025).