Accountability Capture: How Record-Keeping to Support AI Transparency and Accountability (Re)shapes Algorithmic Oversight

Record-keeping for AI accountability has unintended consequences: While documentation practices are meant to enable oversight and transparency, the paper finds that implementing these systems can reshape how organizations operate in ways that weren't anticipated, a phenomenon the authors call "accountability capture."

Employee resistance to documentation requirements is widespread: The survey of 100 practitioners revealed that workers often resist record-keeping practices imposed through accountability frameworks, suggesting that transparency requirements may face practical implementation challenges in real-world settings.

Tensions exist between internal and external accountability demands: Organizations struggle to balance what they document for internal purposes versus what regulators and external parties require, creating conflicting record-keeping practices that complicate genuine oversight.

Documentation practices can increase surveillance and privacy risks: The paper highlights that the very systems designed to make AI more accountable may introduce new concerns around employee surveillance, privacy, and data protection—potentially creating new harms while attempting to prevent others.

Policymakers need to consider implementation realities: The authors argue that those designing AI accountability regulations should pay greater attention to how record-keeping requirements actually function in practice, not just their theoretical benefits, to avoid unintended negative effects on organizations and workers.