The democratic process works best when every eligible voter can take part. But too often, people can’t access information about how to register, where to vote, and what candidates stand for in a language and format they understand. For many citizens who are visually impaired, have low literacy levels, or aren’t native speakers, traditional election emails and printed leaflets are simply inaccessible.
In the face of these communication barriers, voice AI can be an incredibly powerful tool for democracy. It can read information aloud for the visually impaired, answer questions in a citizen’s preferred language or dialect, and turn dense official documents into a spoken conversation. But like every new technology, these same capabilities could be misused by bad actors to deceive or misinform voters.
2026 is a significant election year, with national, regional and local elections taking place across every region of the world, including major elections in the United States and Brazil. It's also one of the first election years in which expressive, natural-sounding voice AI is widely available. Below we share how we’re putting voice AI to work to strengthen the democratic process, while preventing and detecting misuse of this technology.
Strengthening the democratic process
Voice AI is already playing a critical role in elections by expanding access to official information from trusted institutions – and Brazil is leading the way globally.
Ahead of its elections in October, Brazil’s Superior Electoral Court (TSE) is adding a voice layer to its ‘Electoral Assistant’, so citizens can use their voices to get information from the TSE’s website, app, and WhatsApp channels. Powered by our text-to-speech and speech-to-text technology, this new voice interface is designed to help counter election misinformation and broaden access to reliable information for all Brazilian citizens. We formalized the partnership through an MoU with the TSE – our first with a national electoral authority – and joined the TSE's Permanent Program to Combat Electoral Misinformation.
We are also working with Brazilian organizations using voice AI to support efforts to verify information. For nearly a decade, Comprova – a coalition of 42 Brazilian media organizations led by the Brazilian Association of Investigative Journalism (Abraji) – has worked tirelessly to identify manipulated content. We will train Comprova partners to use our AI Speech Classifier and other detection tools as part of their verification workflows.
In the United States, we’re working with policymakers, industry bodies and leading technology companies to protect the mid-term elections. We collaborate with fellow members of California’s Countering Tech-Enabled Fraud Task Force to counter voice-enabled scams. We partner with Google DeepMind to watermark our audio so it can be identified as AI generated, and deepfake detection company Reality Defender to strengthen their real-time detection tools which are used by large enterprises and governments. Through ElevenLabs for Government, we're helping governments and civil society deliver always-on citizen services in multiple languages, and we're working to bring this capability to civic institutions in the United States as the 2026 elections approach.
Preventing and detecting misuse of voice AI
As well as deploying our technology to expand democratic access, we’re investing in the safeguards that combat misuse by bad actors. We take a comprehensive, multi-layered approach to safety, with overlapping protections that span our models and products, so that if one layer is evaded, others remain. Some of these safeguards include:
- No Go Voices. Audio samples uploaded to create clones are screened against our proprietary "No Go Voices" (NGV) safeguard, which is designed to block the creation of voices belonging to high-risk individuals such as public officials, political candidates, celebrities, and other widely recognized figures. We continually refine and expand NGV based on internal monitoring, current events, user reports, and external intelligence – including as new election cycles begin.
- Voice CAPTCHA. Voice CAPTCHA compares the voice in an uploaded sample against a real-time recording of the user reading a dynamically generated script.
- Real-time and asynchronous moderation. We use automated systems to help detect content that may violate our policies, including child sexual exploitation and abuse, hate speech, and scams and fraud. These systems can block certain content before it’s created. Generated content may also be automatically classified to support human review, investigation, and enforcement.
- Enforcement. We take appropriate action against users who violate our policies, attempt to circumvent our safeguards, or engage in activity that creates a risk of serious harm.
- Traceability and transparency. Our systems are designed to trace generated content back to the account that created it. Since 2023, we have also offered a free AI Speech Classifier, which allows anyone to check whether an audio sample contains AI-generated content created on ElevenLabs. Recently, we began partnering with Google DeepMind to bring its SynthID watermarking technology to content generated on ElevenLabs so that it can be identified as AI generated using our free Audio Detector.
Our continued commitment
We’re firm believers in the immense benefits of voice AI in public life, and we’re determined to ensure our technology is used for good. We’ll keep hardening our safeguards, learning from the partnerships we have around the world, and adapting our approach as new risks and opportunities arise. We’re glad to be working closely with government and civil society organizations globally, and we look forward to deepening our partnerships through this election cycle and beyond.
