--- title: We Ran Fable 5 Before and After the Ban. Here’s How... description: Kilo ran their KiloBench benchmark suite against Claude Fable 5 both before and after the US government-ordered takedown and retraining. Before the ban (June... canonical: https://daily.dev/posts/we-ran-fable-5-before-and-after-the-ban-here-s-how-often-you-actually-get-fable--j0w9o0cqd twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: We Ran Fable 5 Before and After the Ban. Here’s How Often You Actually Get Fable. | daily.dev og:description: Kilo ran their KiloBench benchmark suite against Claude Fable 5 both before and after the US government-ordered takedown and retraining. Before the ban (June... og:url: https://daily.dev/posts/we-ran-fable-5-before-and-after-the-ban-here-s-how-often-you-actually-get-fable--j0w9o0cqd og:image: https://api.daily.dev/og/posts/J0w9o0CQD.png og:image:alt: We Ran Fable 5 Before and After the Ban. Here’s How Often You Actually Get Fable. og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # We Ran Fable 5 Before and After the Ban. Here’s How Often You Actually Get Fable. **[Kilo Blog](https://daily.dev/sources/kilo-ai-blog)** · 5 min read · 0 upvotes · 0 comments ## Summary Kilo ran their KiloBench benchmark suite against Claude Fable 5 both before and after the US government-ordered takedown and retraining. Before the ban (June 9), 88 of 445 sessions hit interventions — all hard refusals, with 76 ending after one step. After the return (July 4), 98 sessions were flagged, but the nature shifted dramatically: only 17 were refusals while 92 involved silent fallbacks to Opus 4.8, with 861 of 5,408 total steps served by Opus. In 53 sessions, every single step routed to Opus despite the user requesting Fable. The post highlights that silent fallbacks are more dangerous than refusals because outputs look plausible while a different model did the work, affecting costs, evals, and agent reliability. It also marks a precedent: a frontier model pulled by government order and returned with negotiated, classifier-enforced behavior changes. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What changed in Claude Fable 5's safety classifier after it was pulled from the market and relaunched on July 1? The retrained classifier shifted from outright refusals to silent fallback: instead of refusing a flagged request, it now routes the task to Claude Opus 4.8 without the user necessarily noticing. Benchmark testing found only 17 of 98 flagged sessions were refusals post-relaunch, versus 88 of 88 flagged sessions being refusals before the ban, with about one in six steps overall served by Opus instead of Fable. _Teams debugging inconsistent agent output can track which model actually ran each step via daily.dev's coverage of model routing changes._ ### Why was Claude Fable 5 temporarily taken offline by the US government? Amazon researchers reported a technique that bypassed Fable 5's safety classifiers and got it to identify software vulnerabilities, prompting the Commerce Department to issue an export control directive. Anthropic pulled both Fable 5 and Mythos 5 entirely for about three weeks, marking the first time a frontier model was removed from market by a government and returned with negotiated behavior changes. _Anyone assessing frontier model reliability should follow how regulatory intervention reshapes safety guardrails on daily.dev._ ### How often does Claude Fable 5 hand off requests to Opus 4.8 instead of answering directly? After the retrained classifier launched, roughly one in six steps (861 of 5,408) across benchmark sessions were served by Opus 4.8 rather than Fable, and 53 sessions were routed to Opus for every single step, including runs of 25 to 31 steps. Before the ban, zero steps fell back to Opus; every flagged session was simply refused instead. _Developers deciding whether to pin agent workflows to a specific model can weigh fallback rates like these on daily.dev._ ## Similar posts on daily.dev - [I recreated Fable 5 with Opus and agent loops, and it's close enough that I stopped missing the banned model](https://daily.dev/posts/i-recreated-fable-5-with-opus-and-agent-loops-and-it-s-close-enough-that-i-stopped-missing-the-bann-8iuunavyj) · XDA Developers · 7 upvotes · 0 comments - [Why Claude Fable 5 Keeps Silently Switching You to Opus 4.8 — and What Actually Triggers It](https://daily.dev/posts/why-claude-fable-5-keeps-silently-switching-you-to-opus-4-8-and-what-actually-triggers-it-x1ljdp3rs) · Medium · 0 upvotes · 0 comments - [Claude Opus 5.5 vs. Fable 5.1: One overthinks, the other cuts corners](https://daily.dev/posts/claude-opus-5-5-vs-fable-5-1-one-overthinks-the-other-cuts-corners-3o7fx76eo) · The New Stack · 1 upvotes · 0 comments --- Tags: [#ai](https://daily.dev/tags/ai), [#claude](https://daily.dev/tags/claude), [#anthropic](https://daily.dev/tags/anthropic), [#ai-safety](https://daily.dev/tags/ai-safety) [View this post on daily.dev](https://daily.dev/posts/we-ran-fable-5-before-and-after-the-ban-here-s-how-often-you-actually-get-fable--j0w9o0cqd) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"We Ran Fable 5 Before and After the Ban. Here’s How Often You Actually Get Fable.","url":"https://daily.dev/posts/we-ran-fable-5-before-and-after-the-ban-here-s-how-often-you-actually-get-fable--j0w9o0cqd","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/we-ran-fable-5-before-and-after-the-ban-here-s-how-often-you-actually-get-fable--j0w9o0cqd"},"datePublished":"2026-07-10T22:31:08.257Z","dateModified":"2026-09-14T08:39:56.022Z","description":"Kilo ran their KiloBench benchmark suite against Claude Fable 5 both before and after the US government-ordered takedown and retraining. Before the ban (June...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/0a75fa40d84fdd29ba2caf0343082b16?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/0a75fa40d84fdd29ba2caf0343082b16?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Kilo Blog","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Kilo Blog","logo":"https://media.daily.dev/image/upload/s--x3pmrf8D--/f_auto,q_auto/v1774964071/logos/kilo-ai-blog?_a=BAMAMiWQ0","url":"https://daily.dev/sources/kilo-ai-blog"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/we-ran-fable-5-before-and-after-the-ban-here-s-how-often-you-actually-get-fable--j0w9o0cqd","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai,claude,anthropic,ai-safety","timeRequired":"PT5M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Kilo Blog","item":"https://daily.dev/sources/kilo-ai-blog"},{"@type":"ListItem","position":3,"name":"We Ran Fable 5 Before and After the Ban. Here’s How Often You Actually Get Fable."}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/we-ran-fable-5-before-and-after-the-ban-here-s-how-often-you-actually-get-fable--j0w9o0cqd#faq","mainEntity":[{"@type":"Question","name":"What changed in Claude Fable 5's safety classifier after it was pulled from the market and relaunched on July 1?","acceptedAnswer":{"@type":"Answer","text":"The retrained classifier shifted from outright refusals to silent fallback: instead of refusing a flagged request, it now routes the task to Claude Opus 4.8 without the user necessarily noticing. Benchmark testing found only 17 of 98 flagged sessions were refusals post-relaunch, versus 88 of 88 flagged sessions being refusals before the ban, with about one in six steps overall served by Opus instead of Fable. Teams debugging inconsistent agent output can track which model actually ran each step via daily.dev's coverage of model routing changes."}},{"@type":"Question","name":"Why was Claude Fable 5 temporarily taken offline by the US government?","acceptedAnswer":{"@type":"Answer","text":"Amazon researchers reported a technique that bypassed Fable 5's safety classifiers and got it to identify software vulnerabilities, prompting the Commerce Department to issue an export control directive. Anthropic pulled both Fable 5 and Mythos 5 entirely for about three weeks, marking the first time a frontier model was removed from market by a government and returned with negotiated behavior changes. Anyone assessing frontier model reliability should follow how regulatory intervention reshapes safety guardrails on daily.dev."}},{"@type":"Question","name":"How often does Claude Fable 5 hand off requests to Opus 4.8 instead of answering directly?","acceptedAnswer":{"@type":"Answer","text":"After the retrained classifier launched, roughly one in six steps (861 of 5,408) across benchmark sessions were served by Opus 4.8 rather than Fable, and 53 sessions were routed to Opus for every single step, including runs of 25 to 31 steps. Before the ban, zero steps fell back to Opus; every flagged session was simply refused instead. Developers deciding whether to pin agent workflows to a specific model can weigh fallback rates like these on daily.dev."}}]} ```