Sixty organisations from across the global artificial intelligence ecosystem have launched a five-year initiative aimed at enabling an estimated 3.4 billion people to use AI tools in the languages and voices they use every day.
The coalition, announced on September 21, brings together frontier AI laboratories, researchers, governments, technology developers, philanthropic organisations, community groups and implementation partners. The signatories say they will pool expertise and resources to address one of the most persistent barriers to equitable access to artificial intelligence: the limited support for thousands of languages that remain underrepresented in major AI systems.
The initiative is built around a shared objective of expanding access to safe, effective and culturally relevant AI services for speakers of languages that currently receive limited support from widely used models.
While AI tools have advanced rapidly in a small group of widely spoken languages, much of the world remains poorly served. Of the roughly 7,000 languages spoken globally, only a limited number have the large, high-quality datasets, technical tools and evaluation systems needed to develop reliable AI capabilities.
As a result, people who communicate in so-called low-resource languages can encounter AI systems that are less accurate, less useful or unable to understand local expressions, speech patterns and cultural context.
The coalition said the issue is not simply one of translating text. Voice-based tools will be particularly important in communities where typing, literacy levels, digital interfaces or access to text-based technologies may limit the usefulness of conventional AI applications.
Dialects, slang, idioms and culturally specific expressions can substantially change the meaning of a request or response. In areas such as health care, education, financial services, agriculture and public services, inaccurate interpretation could limit the value of AI tools or, in some cases, create serious risks.
The organisations behind the initiative say better language and voice capabilities could allow billions of people to access AI-enabled information and services in forms that are more natural, relevant and practical for everyday use.
The initiative will focus on four main areas over the next five years. The first is the development of what organisers describe as an open language layer: shared and responsibly governed language-data infrastructure that developers can use under open licences.
A second priority will be measuring progress through assessments and benchmarks designed to show whether language capabilities are improving in meaningful ways, rather than merely expanding the number of languages listed as supported.
The third area will focus on converting language data into working tools, including AI models and applications that can be used by a broad range of developers rather than only the largest technology companies.
The fourth priority is responsible deployment. The coalition said its work will be guided by safeguards relating to privacy, consent, data sovereignty and the protection of the communities whose languages, voices and cultural knowledge are used in building AI systems.
The announcement reflects growing concern that AI could deepen existing digital inequalities if the technology remains designed principally for speakers of high-resource languages. Large language models depend on extensive collections of text, audio and other data, but many languages lack digitised materials, standardised spelling systems, speech recordings, technical research or locally relevant evaluation tools.
For African countries, where linguistic diversity is exceptionally high, the gap is particularly important. AI systems that can understand and communicate effectively in local languages could improve access to agricultural advice, educational content, health information, government services and financial tools. However, advocates have also cautioned that such systems must be developed in partnership with local communities to prevent the misuse of cultural data or the reproduction of harmful errors and biases.
The coalition said it aims to build on years of work by researchers, local organisations, technology developers and public institutions seeking to improve AI performance in underrepresented languages. Rather than replacing existing projects, the new commitment is intended to connect efforts across datasets, models, benchmarks, applications and real-world deployment.
Organisers are inviting additional partners to join the initiative, particularly organisations that can contribute local language expertise, data stewardship practices, funding, research capacity, public policy support or implementation experience.
