AI Policy & Governance, Free Expression
CDT and Cornell Global AI Initiative Call for Meaningful Advancements and Investment into Linguistic Diversity at the First UN Global Dialogue on AI Governance
Also authored by Aditya Vashistha, CDT Non-Resident Fellow & Assistant Professor, Cornell University + Global AI Initiative Lead.
The Center for Democracy & Technology (CDT) and Cornell Global AI Initiative submit these joint comments to shape the first session of the Global Dialogue on AI Governance hosted by the UN. These comments expand on spoken remarks delivered by Kate Ruane, director of CDT’s Free Expression Project, during the first consultation on the Global Dialogue held virtually on March 18, 2026.
CDT is a nonprofit 501(c)(3) organization that works to advance human rights in the digital age. Among our priorities, CDT advocates for the responsible and rights-respecting design, deployment, and use of new technologies including artificial intelligence (AI), and promotes the adoption of robust, technically-informed solutions for the effective regulation and governance of AI systems.
The Cornell Global AI Initiative brings global perspectives into the design, evaluation, and governance of AI technologies. In a short time, the initiative has built strong momentum across the university, bringing together more than 15 faculty across seven units, including Computing and Information Science, Communications, Landscape Architecture, Operations Research, Linguistics, and Science and Technology Studies. This interdisciplinary network examines how AI systems interact with culture and inequality while developing approaches that ensure AI technologies are safe, humanistic, pluralistic, and beneficial to diverse communities globally.
For these comments, we focus exclusively on procedural and participatory tactics to advance the linguistic capabilities of AI systems and ensure safety mitigations work equally well in multilingual and multicultural contexts.
Currently, developers and deployers trying to build systems that work equally well across the world’s 7000 languages face a practical constraint: There is very little high quality data in many of the world’s major languages. While there are now more public AI training datasets representing these historically under-represented languages, an audit conducted by the Data Provenance Initiative in 2024 found that the relative representation of different languages and regions in the data used by large AI models has not changed since 2013.
Beyond data availability issues, the representativeness of existing datasets in many language families is also an issue. Few datasets represent the specific terms and concepts people use in the domains where AI technologies may be deployed, such as when seeking healthcare information. This dearth in availability of high quality representative data limits model performance and capabilities in non-English languages and contexts.
In an effort to bridge this gap, AI developers rely on a range of alternative sources of data, but much of it is imperfect. Synthetic data (that is, data generated using AI tools or English-language data translated using AI translation tools), data provided by government actors, and data in other semantically-similar languages are often used to train systems, but all of these tactics raise both practical and human rights concerns. As the International AI Safety Report outlines, advancements in AI capabilities remain jagged, particularly in languages other than English.
The shortcomings of these and other approaches are scarcely documented and mitigated because AI developers and deployers rarely conduct robust testing and evaluations in languages other than English or do so with imprecise instruments. CDT has outlined shortcomings of automated benchmarking tools broadly and how they should be improved to be more robust and pursue depth rather than simply breadth. In languages other than English, these automated tools used to judge outputs or evaluate systems are often more prone to failure as many languages are scarcely represented in available evaluation tools. Or, evaluation tools are not valid or suitable for the domain in which the AI system will be used. This also limits model developers and deployers from testing how well multilingual safety guardrails work in non-English contexts, making AI governance levers proposed in this and other arenas insufficient if they are easily jailbroken or circumvented in languages other than English.
The Global Dialogue has already committed to ensuring linguistic diversity is part of the global AI governance conversation. The Dialogue must ensure conversations move beyond documenting the “digital language gap” and move towards filling current technical and research gaps. Tangibly, this requires:
Fostering an ecosystem of open systems to enable developers to fine tune and build language-specific tools.
One prevailing method for building high quality AI systems in languages other than English is to fine-tune existing, light-weight open source models with language and domain-specific data. Encouraging developers to make models of different sizes open and available can create a more flourishing ecosystem of language-specific tools by reducing the infrastructural and cost-related barriers to developing a model from scratch, addressing concerns related to data stewardship and sovereignty, and expanding who is able to build a system in their own language. This includes ensuring the availability of open-source benchmarks and evaluation mechanisms, open source software, open data and open artificial intelligence models. This is especially essential for multimodal datasets, such as voice and speech-to-text systems, where high quality data is even more scarce, though some actors, such as Microsoft’s Paza project, are addressing this gap. Already, the Mozilla Data Collective offers a paradigm of this approach by stewarding the biggest collection of datasets in non-English languages that are available to be used to train and fine-tune systems. The Global Dialogue should bring together stakeholders needed to support this ecosystem to plan concrete steps for its expansion across languages.
Establishing multistakeholder channels to incorporate language and domain experts in the development and deployment of AI systems that are linguistically and culturally aligned.
Due to the dearth of existing resources in many non-English languages (datasets, evaluations, and more), participation of native language speakers, subject matter experts, and domain experts is critical to developing multilingual AI systems. One pilot program led by Microsoft Research India has shown that on-the-ground healthcare workers are best positioned to shape the data collection and annotation processes required to test AI systems used in healthcare settings in Indian languages towards more useful and effective performance of their systems. These channels for participation are essential when it comes to AI systems used in high risk settings as Cornell research has shown. The Global Dialogue can elevate these types of examples, identify best practices within them through the application of the NetMundial +10 Sao Paulo Multistakeholder Guidelines, and encourage AI companies to develop channels and use multistakeholder-developed best practices to incorporate those with the most expertise and linguistic fluency into the process when they develop and deploy AI systems.
Creating a repository of independently-created multilingual evaluations or urging existing AISIs to adopt these evaluations to incentivize their use.
As stated above, AI developers and deployers don’t always have access to high quality evaluations in all of the world’s languages. This can be particularly important when usage of a system evolves over time; for example, if a deployer sees an increase in use by Brazilian Portuguese speakers but does not easily have access to expertise in that language and context to test if model safeguards are proving effective. Having independent community-created evaluations to test systems’ performance and safety guardrails in a specific language can identify gaps and prioritize collaboration and development of needed resources . Right now, evaluations developed by native language speakers aren’t easily available or accessible and searching for the right one may be time consuming.
An independent body can equip developers and deployers with appropriate evaluations and set up criteria to enable developers and deployers to choose the right evaluation. ML Commons has sought to fill this role to some degree but lacks the resources to serve as a comprehensive central repository; nonetheless, like other researchers, it has also sought to improve adoption of high quality evaluation tools by both creating a multilingual benchmarking suite and developing guidance to deployers of evaluations on how to best use it. AI red-teaming expert Roya Pakzad has also developed a platform for multilingual evaluations to be used by non-technical deployers and procurers of systems to increase informed selection of AI tools, which is another function a central repository can play.
Some research questions around scale and how to ensure these community-led evaluations drive improvements in AI capabilities remain and as such this body should document and advance relevant research. The Global Dialogue should drive discussions around identifying similar cross-platform gaps and developing independent multistakeholder intermediaries to fill such gaps. Addressing the barriers intermediaries like ML Commons and others face in making high quality evaluations available to developers, and offering best practices on how developers and deployers should create and use evaluations and document and use findings from evaluations are examples of how the Global Dialogue can help scale the adoption of high quality and public-interest governance levers.
Directing funding towards language-specific research networks
AI research groups that lead development of tools and paradigms for building AI systems in the world’s languages such as Masakhane, SEA-LION, IndoNLP, AmericasNLP, and ARBML often have deep expertise about where the largest gaps are in their language’s specific research but are sorely lacking the funds necessary to address them. The Global Dialogue can convene philanthropy, academia, and independent research groups to ensure a level playing field when it comes to research and development of multilingual tools.
What’s at stake when systems do not work equally well in languages other than English? Users may not be able to equally access and benefit from advancements in digital technology, creating what some experts call a “digital linguistic divide” and hampering progress towards SDG goals, including SDG 10 on “Reduced inequalities” and SDG 16 on “peace, justice, and strong institutions”. The latter is relevant here as AI systems are increasingly deployed by governments and other institutions across a range of domains from healthcare settings to being used to determine who receives public services. These systems must serve all people regardless of the language they speak.
Facts Only
The Center for Democracy & Technology (CDT) and Cornell Global AI Initiative submitted joint comments to the UN’s first Global Dialogue on AI Governance.
The comments were delivered during a virtual consultation on March 18, 2026.
CDT is a nonprofit organization advocating for responsible AI design and governance.
The Cornell Global AI Initiative involves over 15 faculty members across seven academic units.
A 2024 audit by the Data Provenance Initiative found no change in language representation in AI training data since 2013.
AI developers often use synthetic data, government-provided data, or semantically similar languages to compensate for data shortages.
Automated benchmarking tools for non-English languages are frequently unreliable or unsuitable for specific domains.
The Global Dialogue has committed to addressing linguistic diversity in AI governance.
Proposed solutions include open-source ecosystems for language-specific AI tools and multistakeholder participation in AI development.
Microsoft Research India’s pilot program involved healthcare workers in data collection for AI systems in Indian languages.
Independent multilingual evaluations are scarce, with platforms like ML Commons and Roya Pakzad’s work attempting to fill gaps.
Research networks like Masakhane, SEA-LION, and IndoNLP lack sufficient funding for language-specific AI development.
Unequal AI performance in non-English languages risks widening global inequalities and hindering Sustainable Development Goals.
Executive Summary
The Center for Democracy & Technology (CDT) and Cornell Global AI Initiative have submitted joint comments to the UN’s first Global Dialogue on AI Governance, emphasizing the need for meaningful advancements in linguistic diversity within AI systems. The submission highlights persistent challenges in AI development, particularly the lack of high-quality data for the world’s 7,000 languages, which has remained unchanged since 2013 despite increased public datasets. Current AI models often rely on imperfect alternatives like synthetic data or translations, raising practical and human rights concerns. Additionally, evaluation tools for non-English languages are frequently inadequate, limiting the effectiveness of safety guardrails in multilingual contexts. The comments propose four key actions: fostering open ecosystems for language-specific AI tools, establishing multistakeholder channels to incorporate linguistic and domain experts, creating repositories of independent multilingual evaluations, and directing funding toward language-specific research networks. The stakes are high, as unequal AI performance across languages risks deepening global inequalities and undermining progress toward Sustainable Development Goals, particularly in governance, healthcare, and public services.
The submission underscores the urgency of moving beyond documenting the "digital language gap" to actively addressing technical and research gaps. Examples like Microsoft’s Paza project and Mozilla’s Data Collective demonstrate progress, but systemic barriers remain. Without intervention, AI systems may fail to serve non-English speakers equitably, exacerbating disparities in access to technology and institutional services.
Full Take
This submission from CDT and Cornell Global AI Initiative presents a compelling case for addressing linguistic diversity in AI governance, but it also reveals deeper systemic challenges in global AI development. The strongest version of this narrative highlights a critical gap: despite growing awareness of the "digital language divide," progress in multilingual AI remains stagnant, with data representation unchanged since 2013. The authors deserve credit for proposing concrete solutions—open ecosystems, multistakeholder collaboration, and funding for language-specific research—rather than merely diagnosing the problem. However, the analysis also exposes a tension between technical feasibility and equitable outcomes. The reliance on synthetic data and translations, while practical, raises ethical concerns about accuracy and cultural alignment. The call for independent evaluations is particularly noteworthy, as it shifts power from AI developers to linguistic communities, but the lack of funding for grassroots networks like Masakhane suggests structural barriers persist.
Patterns detected: none. The narrative avoids manipulation tactics, focusing on evidence-based advocacy. The root cause appears to be a paradigm of AI development that prioritizes scalability and cost-efficiency over linguistic inclusivity. This echoes historical patterns of technological colonialism, where dominant languages and cultures shape global systems, marginalizing others. The implications are profound: without intervention, AI could entrench inequalities, limiting access to services and reinforcing institutional biases. The second-order consequences include potential backlash against AI adoption in non-English contexts, further fragmenting global digital governance.
Bridge questions: How might funding mechanisms be restructured to prioritize language-specific AI research without reinforcing dependency on Western institutions? What role should governments play in mandating multilingual AI standards, and how could this be enforced without stifling innovation? What metrics would best measure progress in closing the digital language divide?
Counterstrike scan: If this were part of a coordinated influence campaign, the playbook might involve exaggerating the urgency of linguistic diversity to push for centralized governance models that benefit specific actors. However, the content aligns with genuine advocacy for equitable AI development, with no signs of manipulation or hidden agendas.
Sentinel — Human
This text is a well-structured piece of policy advocacy that successfully synthesizes technical constraints with multi-stakeholder governance solutions, exhibiting strong human analytical depth.