An election authority adds an AI voice layer to its citizen assistant

ElevenLabs' voice technology integrates with the Brazilian Electoral Court's official assistant. A full-scale test for access to information.

A major election year, 2026 is also one of the first in which expressive synthetic voices are widely accessible. ElevenLabs is releasing an update on its work in this area, balancing the use of voice AI for access to election information with measures against its misuse. The company highlights an accessibility benefit—reading text aloud for a visually impaired person, responding in a citizen's language or dialect, converting a dense official document into a spoken conversation—while acknowledging that these same capabilities can be used to deceive voters.

The most concrete development comes from Brazil. Ahead of its October elections, the Superior Electoral Court is adding a voice layer to its Electoral Assistant, allowing citizens to query its website, app, and WhatsApp channels by voice, powered by ElevenLabs' transcription and synthesis technologies. The company states it has formalized this partnership through a memorandum of understanding, its first with a national electoral authority, and has joined the Court's permanent program against electoral disinformation. Additionally, it plans to train partners of the Comprova fact-checking coalition, which brings together dozens of Brazilian newsrooms, on the use of its speech classifier and other detection tools.

In the United States, ahead of the midterm elections, the company says it is working with policymakers and other tech players, notably within a California task force against fraud, and with the company Reality Defender on real-time detection.

Regarding safeguards, the publisher describes a multi-layered approach. A feature called No Go Voices aims to block the cloning of voices of high-risk individuals, public officials, candidates, or well-known figures. A Voice CAPTCHA compares an imported sample to a live recording read by the user. This is complemented by automated moderation, content traceability back to the account that produced it, a speech classifier open to everyone since 2023, and the integration of Google DeepMind's SynthID watermarking to identify audio as generated.