Do Responsible AI Artifacts Advance Stakeholder Goals? Four Key Barriers Perceived by Legal and Civil Stakeholders

RAI artifacts may create a false sense of accountability: While tools like Model Cards and Transparency Notes are intended to document AI development processes, legal and civil society stakeholders worry these documents could be misused to demonstrate compliance without meaningfully addressing underlying harms—potentially shifting power away from external oversight bodies toward technology companies.

Accessibility and usability gaps limit external stakeholder participation: The study found that government, legal, and civil society actors struggle to effectively use RAI artifacts due to technical complexity and design choices that prioritize internal company use. This creates barriers for those tasked with auditing and governing AI systems, potentially concentrating power among those with technical expertise.

RAI artifacts may inadequately represent affected communities: Participants expressed concern that these artifacts focus on documenting technical processes rather than centering the voices and experiences of downstream end-users and marginalized communities most impacted by AI systems, potentially reinforcing existing power imbalances.

Regulatory reliance on RAI artifacts without proper enforcement mechanisms is problematic: As regulations like the EU AI Act increasingly mandate these artifacts for compliance, stakeholders worry that without clear enforcement standards and consequences for inadequate documentation, companies could use RAI artifacts as "compliance theater" rather than genuine governance tools.