In this video, Andrew from Box demonstrates Anthropic’s Claude Fable 5 on enterprise content in Box AI. The demo highlights two realistic knowledge-worker workflows: reviewing a vendor agreement against its exhibits, and comparing an earnings narrative against the underlying financial data. Fable 5 produces structured, source-grounded analysis — identifying exhibit-only obligations, conflicting SLA terms, and places where management commentary downplays or omits important financial context. What you'll learn in this video Model Behavior is Box's ongoing series where we test the latest AI models on the kinds of files enterprise teams actually work with — contracts, earnings reports, exhibits, and financial tables. In this episode, we put Claude Fable 5 (Anthropic's newest model, built for demanding reasoning and long-context work) through two realistic enterprise workflows: can it connect the dots across documents, catch inconsistencies, and surface what a human reviewer should pay attention to? Demo 1 — Contract review: main agreement vs. exhibit A The test uses a fictional vendor agreement between Data Pipeline and Meridian Financial Group. Claude Fable 5 is asked to identify obligations in exhibit A not referenced in the main agreement and flag any conflicting terms. The model surfaces exhibit-only obligations — dedicated engineering resources for the first 90 days, plus specific support response times (P1: 1 hour, P2: 4 hours, P3: 24 hours) — and flags a direct conflict: clause 3.1 specifies 99.5% uptime for all production services, while exhibit A specifies 99.9% for the real-time analytics API. It also notes that exhibit A is referenced but not fully included, and exhibit B is missing entirely. Demo 2 — Earnings review: does the narrative match the numbers? The second test uses an internal Q2 earnings review for a fictional company called Novatek. Claude Fable 5 is asked to compare the CEO letter to the actual financial tables and identify where the narrative downplays or contradicts the data. It confirms what holds up — cloud and SaaS up 24% YoY, above guidance, 44% of revenue — then identifies the gaps: the EMEA restructuring charge's contribution to margin pressure is absent from the narrative, hardware came in 8% below guidance, enterprise software missed guidance, and gross margin declined 180 basis points. The broader takeaway Claude Fable 5 is especially useful for judgment-heavy knowledge work — where important details are spread across tables, exhibits, and fine print, and the real question is whether the pieces are consistent with each other. With Box AI, it's already working on the content you store in Box. FAQs Q: What is Model Behavior by Box? A: A Box video series that tests the latest AI models on realistic enterprise workflows — contracts, earnings reports, exhibits, and financial tables. Q: What did Claude Fable 5 do in this video? A: Two tasks: (1) compared a vendor contract's main agreement to exhibit A to surface missing obligations and flag conflicting terms, and (2) compared a CEO earnings narrative to the underlying financial tables to identify where the narrative contradicted the data. Q: What contract conflict did Claude Fable 5 catch? A: A direct conflict between clause 3.1 (99.5% uptime for all production services) and exhibit A (99.9% uptime for the real-time analytics API), plus exhibit-only obligations with no counterpart in the main agreement. Q: What did Claude Fable 5 find in the earnings review? A: The EMEA restructuring charge's contribution to margin pressure was absent from the narrative; hardware came in 8% below guidance; enterprise software missed guidance; gross margin declined 180 basis points. Q: Which teams benefit most? A: Legal, procurement, finance, investor relations, and operations — any team that cross-references documents or verifies that a narrative aligns with underlying data. Q: Does this require moving content out of Box? A: No. Claude Fable 5 works directly on the content already stored in Box.

 161      0