SpeechifyAI and our elections

Synthetic voices are now part of how elections are contested, for good and for ill. This page sets out the safeguards SpeechifyAI operates to prevent our voice technology being used for election misinformation.

Voice AI and the democratic process

Speechify began as an accessibility company, and that is still the lens we see this technology through. Text to speech puts civic information within reach of people who would otherwise struggle with it: voters with dyslexia or low vision, voters reading in a second language, voters who simply absorb information better by ear. Synthetic voice lowers the cost of delivering accurate election information to every one of them, in their own language, at scale.

The same properties lower the cost of deceiving them. In January 2024, a robocall reached an estimated 5,000 to 25,000 New Hampshire voters using a cloned voice of then-President Biden, telling them not to vote in the primary. The FCC proposed a $6 million fine over the call. What the incident demonstrated was not sophistication but cheapness: a scraped audio sample, a cloning API, and a dialler were enough to impersonate a head of state to thousands of voters.

Voice AI providers sit at the start of that pipeline. We believe the obligation to keep it out of that pipeline sits with us as much as with anyone downstream, and the measures on this page are how we act on that.

What does the law require of synthetic voices?

Regulation of AI voice technology has advanced quickly, and we treat it as a floor rather than a target.

In February 2024, the FCC ruled unanimously that AI-generated voices are “artificial” under the Telephone Consumer Protection Act. The ruling covers calls made with any current voice generation technology, ours included: such calls require the called party’s prior express consent, in writing for telemarketing.

Article 50 of the EU AI Act has applied since 2 August 2026. It requires providers to mark synthetic audio in a machine-readable form, and requires deployers to label AI-generated audio that convincingly resembles a real person. Systems already on the market before that date have a transitional window for the marking requirement, currently set at 2 December 2026 under the AI Omnibus, and we will meet our marking obligation within it. Penalties reach €15 million or 3% of global annual turnover, whichever is higher.

Under GDPR, a recording of a person’s voice is their personal data. The speaker has the right to know it is being processed, to object, and to have it erased. Any service that accepts voice samples carries those obligations, and they long predate the AI Act.

This page is a statement of our measures, not legal advice. Organisations serving the EU, the UK, or a US state with its own synthetic media law should take their own counsel on what applies to them.

Preventing misuse: the safeguards we operate today

We take a layered approach: consent is verified before a voice can be cloned, the evidence is retained, and everything a customer creates is attributable.

Verified consent for voice cloning

Voice cloning on the SpeechifyAI API is moving to verified consent. The platform issues a single-use challenge phrase and requires a fresh recording of the speaker reading it; the recording is transcribed and matched against both the phrase and the voice being cloned, then retained as the consent record for the voice.

The earlier flow, which accepted a written assertion of consent, is deprecated and is being retired ahead of the November midterm elections, on a migration window deliberately shorter than our standard 12 months. Affected workspaces will get the switch-off date ahead of time. Details are in the announcement and the Voice Cloning API guide.

Retained consent records

Every cloned voice carries the consent evidence supplied at its creation; for voices created on the verified flow, that includes the recording itself. If a voice is later disputed, the question is settled against that record.

Workspace-level traceability

Every cloned voice is owned by the workspace that created it, and every generation request is logged against the API credentials that made it, so misuse traces to an account. We act on violations of our terms, including removing voices and terminating access.

Working across the industry

No single provider can protect an election. A bad actor refused by one platform must meet the same standard at the next, which is why we consider it good news that our competitors are investing in the same direction. ElevenLabs published its election safeguards in July 2026, including screening uploads against the voices of public figures, a voice CAPTCHA that verifies a live speaker, and watermarking work with Google DeepMind. Their voice CAPTCHA and our consent challenge reach the same conclusion independently: the person behind a voice must be present when it is cloned.

That convergence matters more than any single feature. When speaker verification becomes the industry floor, the cheap attack stops being cheap.

Our continued commitment

The measures above are a beginning, not an end. We are expanding our election-integrity work across detection, provenance, and abuse prevention, including the EU AI Act’s machine-readable marking obligation within its transitional window for systems already on the market.

If you are an election official, a researcher, or a journalist investigating suspected misuse of synthetic voices, contact us at support@speechify.com and we will prioritise it. If you believe a Speechify voice has been used to impersonate someone without consent, the same address reaches the team that can act on it.

Elections depend on voters trusting what they hear. We intend for our technology to strengthen that trust through this election cycle and beyond.

FAQ

Is AI voice cloning legal?

Cloning a voice with the speaker’s consent is legal in most jurisdictions. What regulators target is deception: US robocalls using AI voices require the called party’s prior express consent under the TCPA, and the EU AI Act requires synthetic audio to be marked and convincing impersonations to be labelled. Consent and disclosure are the dividing line, which is why SpeechifyAI verifies consent at the point of creation.

How does Speechify verify consent for a voice clone?

The API issues a single-use challenge phrase through POST /v1/voices/consent-challenges. The speaker reads the phrase aloud, the recording is submitted with the create call, and Speechify transcribes it, matches it against the issued phrase, and retains it as the consent record for the voice. A create request without a verified recording is rejected.

What does the EU AI Act require for synthetic audio?

Article 50, in application since 2 August 2026, requires providers to mark AI-generated audio in a machine-readable form and requires deployers to disclose deepfakes that would appear authentic to a listener. Systems already on the market before that date have a transitional marking window, currently set at 2 December 2026 under the AI Omnibus. Penalties reach €15 million or 3% of global annual turnover, whichever is higher.

How do I report misuse of a Speechify voice?

Email support@speechify.com with the audio or a link to where it appears and any context you have. Reports of election-related misuse or unauthorised impersonation are prioritised, and every cloned voice on the platform is attributable to the workspace that created it, so verified reports lead to enforcement.