Terms of Care: Designing for Participation in Data Governance by Disability Communities

Madiha Zahrah Choksi, Reeda Shimaz Huda, Rita Faia Marques, Anja Thieme, Cecily MorrisonarXiv 2026 · arXiv

TL;DR

This paper introduces a modular licensing framework that empowers disability advocacy organizations to participate in AI data governance. By decomposing standard Creative Commons licenses into adaptable modules, the researchers demonstrate how "high-context" communities can translate their relational values and histories of harm into enforceable terms for sharing AI training data.

What Problem Does This Paper Address?

AI-generated media often fails to represent disability accurately, frequently falling into tropes of pity, ridicule, or complete erasure. While the obvious solution is to include more community-specific data in training sets, doing so creates a "participation paradox." Sharing sensitive community data openly exposes these groups to decontextualization and exploitation—such as the recent trend of using AI to create fictitious influencers with Down syndrome to sell sexual content.

Existing data governance tools, like standard open-source licenses (MIT, Apache, CC0), are often "one-size-fits-all" and fail to account for the nuanced, relational norms of disability communities. The authors characterize these as high-context communities, where decision-making relies heavily on shared history, trust, and implicit norms rather than just formal, documented rules. There is a critical gap in providing these communities with tools that allow them to share data for "good representation" while maintaining control over how that data is used.

Methods

The researchers conducted a two-phase study with three disability advocacy organizations (representing people with dwarfism and vision impairments):

  1. Phase 1: Organizational Governance Interviews: Semi-structured interviews to map existing governance practices, organizational memory, and histories of data-related harm.
  2. Phase 2: Modular License Design & Evaluation: An interactive webinar followed by a hands-on activity where participants used a "license-building worksheet."

The core innovation is the Modular License Decision Flow. Instead of choosing a static license, participants selected a base (Permissive CC0 or Copyleft CC BY-SA) and then layered specific modules to reflect their values.

![Fig.

  1. Modular License Decision Flow. Participants selected a base Creative Commons license type and layered modular clauses (Attribution, Allowed Uses, Restrictions, Share-Alike, Community Values) to produce a tailored license statement.](/figures/art_6dce1006e5e24a8b977de3697884ec86)

Datasets

While the paper focuses on governance, it builds on prior work where these three organizations curated community-centered image datasets. Each dataset consists of 400 photos depicting community members in diverse, everyday activities, complete with structured annotations and metadata specifically formatted for AI training and evaluation.

Key Insights and Ideas

  • Governance as Organizational Memory: Governance isn't just about rules; it's shaped by "situated harms." For example, one organization’s caution was rooted in a past incident where a third-party contractor bypassed their protocols to contact families directly.
  • Relational Accountability: In high-context communities, trust is built through ongoing communication rather than a one-time consent form. The choice of communication channel (e.g., a transparent WhatsApp group vs. private messages) is itself a governance act.
  • Licensing as a "Boundary Object": The modular license acted as a bridge, allowing non-experts to translate abstract legal concepts into concrete community protections.
  • The Rejection of "Permissiveness": All three organizations rejected the CC0 (public domain) model. They viewed "Copyleft" (CC BY-SA) as essential for maintaining community ownership and ensuring that any improvements to the data were shared back under the same terms.

Key Results

The study resulted in three bespoke licenses that reflected specific community "red lines":

  • Sector-Specific Bans: One organization representing people with dwarfism explicitly prohibited data use in the pharmaceutical sector (to oppose "cure-oriented" messaging) and the entertainment sector (to prevent stereotypes related to circus performances or wrestling).
  • Contextual Permissions: Another organization allowed images for newspaper reporting but strictly prohibited their use on social media to prevent the creation of derogatory memes.
  • Value Statements: Participants successfully embedded explicit ethical commitments—such as "disability justice" and "anti-surveillance"—directly into the license text.
  • Educational Impact: Survey results showed that the modular approach significantly improved the participants' confidence in acting as "data stewards" for their communities.

Limitations and Future Work

The authors acknowledge that licensing is not a silver bullet. While a license defines the terms of care, it does not provide the technical infrastructure for monitoring or enforcement. A bad actor can still scrape data and ignore the license.

Future work is needed to:

  1. Develop platform-level tools (e.g., on Hugging Face or GitHub) that support these guided, modular license builders.
  2. Create technical mechanisms for tracking data compliance and enabling community-led "takedown" requests.
  3. Scale the approach to other marginalized groups who face similar high-context governance challenges.

Summary updated

Abstract

Improving disability representation in AI-generated media requires community-specific datasets, yet openly sharing such data can expose communities to misuse, decontextualization, and exploitation. We examine how disability communities can participate in dataset governance through context-sensitive licensing. Working with three disability advocacy organizations, we conducted semi-structured interviews, an educational webinar and modular license activity to promote community participation in data stewardship. Our findings show that governance in these high-context communities relies on organizational memory (especially situated harms); both formal authority and informal relational practices; communication; and resulting trust. The modular license activity made organizational data choices more accessible and values-led, allowing communities to adapt Creative Commons clauses to express permitted uses, sector-specific restrictions, safeguarding requirements, and commitments to respectful representation. Yet, concerns remain about monitoring and enforcing appropriate data use. We contribute (1) an empirical account of governance tensions, (2) a practical modular approach, and (3) design implications for AI data platforms that support relational, context-sensitive governance beyond standardized, one-size-fits-all frameworks.
BibTeX
@misc{choksi2026terms,
  author = {Madiha Zahrah Choksi and Reeda Shimaz Huda and Rita Faia Marques and Anja Thieme and Cecily Morrison},
  title = {{Terms of Care: Designing for Participation in Data Governance by Disability Communities}},
  year = {2026},
  eprint = {2611.15720},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2611.15720},
}
Figure