CDT urges UN rapporteur to put AI at the center of free expression work

AI companies and human rights commitments

CDT says the mandate has an opportunity to shape the governance of AI and other emerging technologies, because their development bears directly on the flow of information and ideas, which it ties to Article 19. According to the organization, the first step is securing meaningful commitments from AI companies to human rights principles and frameworks.

The submission points to several existing instruments: UNGA Resolution 78/265 on trustworthy AI, the Freedom Online Coalition's Joint Statement on AI and Human Rights, the OECD Principles for Trustworthy AI, and a foundational paper from the UN Office of the High Commissioner for Human Rights on responsible development and deployment of generative AI. CDT says these principles can and should govern AI systems in ways that improve transparency and accountability throughout the development lifecycle.

Moderation tools and the language gap

CDT also draws attention to large language models (LLMs) that online services, including AI companies, use to moderate content and mediate access to information. The group says such tools and safeguards are meant to keep outputs safe, useful and relevant, but can harm users' rights. It cites emerging research indicating that AI-powered safeguards can suppress speech related to marginalized identities and politics, especially when user queries or behavior do not conform to the views of those in power.

It suggests the Special Rapporteur could pursue a thematic report or multistakeholder consultations, including with technical experts, and help companies apply international human rights principles when building LLM-based tools.

A second concern is language. CDT warns of a two-tiered information environment, one for English and one for the rest of the world. It cites the most recent International AI Safety Report, which finds that multilingual capabilities and reliability across contexts remain "jagged." According to the submission, systems are prone to hallucinate more, cite less reputable sources and bypass safety guardrails in languages other than English.

The group notes that AI companies are increasingly committing to localize responses based on local law and culture. CDT says this can be beneficial in some cases, but in rights-repressive regimes it may cost users access to information on issues such as LGBTQ+ identity and rights, women's rights and political dissent.

Participation gaps in AI governance

CDT describes a significant asymmetry in who takes part in AI governance. Frontier AI models are developed primarily by private companies based in a small number of countries. The UN High-level Advisory Body on Artificial Intelligence reported that 118 countries, primarily in the Global South, are not participants in seven major non-UN international AI governance initiatives. CDT adds that private investment outpaces governance, and many governments lack the technical expertise or legal capacity to act independently. Civil society, it says, faces similar imbalances and cites technical and algorithmic opacity as an obstacle.

The submission also highlights a gender gap. It states that men hold 8 in 10 AI jobs and nearly 9 in 10 AI leadership roles. CDT argues this leaves efforts against tech-facilitated gender-based violence and algorithmic bias without the voices of those most affected. It points to non-consensual intimate imagery, including AI-generated deepfakes, which it says disproportionately affects women and LGBTQ+ people and is used to harm and silence victims. The resulting chilling effect causes women to self-censor, particularly women of color, journalists, human rights defenders and politicians.

What CDT wants the mandate to watch

The submission goes on to name three emerging trends where CDT says the rapporteur's attention would inform rights-respecting governance. The first is an AI-mediated information environment, which CDT describes as a transformative shift in access to and enjoyment of freedom of expression and opinion. People increasingly use AI to get information, both directly through chatbots, AI browsers and agents, and indirectly. The text provided cuts off before the other two trends are described.

For the UN mandate, CDT's message is that AI governance, inclusive participation and company accountability should sit alongside its existing priorities on information quality and public participation.