Fable 5 is back globally: the redeployment routing checklist I run before July 7

Introduction

Eighteen days offline is long enough for every team to pin Opus 4.8 as fallback and forget why Fable was in the routing table in the first place. Fable 5 is back. The question is not "can we call it again." It is "should this workflow pay frontier prices and accept safeguard routing now that Sonnet 5 exists at $2/$10 intro pricing."

I wrote the export-control governance checklist during the suspension and the launch routing guide the week Fable shipped. This post is the redeployment pass: what changed in Anthropic's June 30 statement, what July 7 means for subscription economics, and how to re-promote without repeating June 12's overnight config scramble.

Timeline: suspension to redeployment

• June 9: Fable 5 and Mythos 5 launch globally.
• June 12: US export-control directive; Anthropic suspends both models worldwide.
• June 26: Mythos 5 restored for approved US Glasswing organizations.
• June 30: Export controls lifted; Anthropic publishes redeployment plan and improved classifier.
• July 1: Fable 5 globally available again on Platform, Claude.ai, Code, Cowork.
• July 7: End of 50% weekly usage inclusion on subscription plans; credits required after.

What changed in safeguards (and why your eval suite matters)

Anthropic confirmed the Amazon-reported jailbreak was a borderline defensive cyber case, not unique Mythos-level offensive capability. Less capable models including Opus 4.8 and GPT-5.5 identified the same vulnerabilities in their testing. The redeployment ships an improved classifier that blocks the specific technique in over 99% of cases, with user notification and automatic routing to Opus 4.8 when blocked.

The tradeoff is false positives on benign coding and debugging tasks touching security-adjacent code. That is not a reason to avoid Fable. It is a reason to measure cost per completed task with safeguard routing included, not list price per token. Same lesson as the Fable launch routing guide: safeguards silently shift spend to Opus.

Anthropic also outlined a Glasswing industry framework for jailbreak severity (capability gain, breadth, ease of weaponization, discoverability) with Amazon, Microsoft, and Google. Treat that as a signal that frontier launches will face faster government and industry triage. Your routing table needs revocable IDs and evals, not permanent defaults.

The redeployment checklist I run this week

1. 01
   Confirm July 7 billing cliff with finance

   50% weekly inclusion on Pro/Max/Team ends July 7. Standard Enterprise seats without credits still cannot access Fable. Premium Enterprise seats include Fable through July 7 then need credits enabled.
2. 02
   Re-run cost-per-completed-task evals vs Sonnet 5

   Sonnet 5 shipped June 30 at intro pricing and beats Opus 4.8 on some agentic benchmarks. Fable must win on your tasks, not on leaderboard averages. See Sonnet 5 migration checklist.
3. 03
   Keep Opus 4.8 as explicit fallback in config

   Safeguard routing sends blocked requests to Opus 4.8. Your fallback chain should expect that path and log model_id plus stop_reason on every run.
4. 04
   Test cyber-adjacent prompts in staging

   Run representative security debugging prompts and measure refusal rate and false positives before promoting Fable on developer-facing defaults.
5. 05
   Log model_id and refusal events in observability

   Wire the agent observability stack before redeployment, not after finance asks why Opus spend spiked.

Common redeployment mistakes

• Re-pinning fable-5 everywhere because "it's back" without evals.
• Ignoring July 7 subscription economics when planning Q3 budgets.
• Assuming zero-retention enterprise agreements cover Fable traffic (30-day retention still applies).
• Skipping Sonnet 5 comparison because Sonnet feels "good enough from June."
• No documented owner for the next export-control or safeguard incident.

Conclusion

Fable 5's return is not a reset to June 9. You now have Sonnet 5 as a stronger mid-tier, a proven export-control revocation path, and a classifier that trades some false positives for jailbreak coverage. Re-promote Fable where cost per completed task wins after safeguard routing. Keep fallbacks pinned. Treat July 7 as a billing event. The teams that handle redeployment well treat it like any other model migration with evals, not like turning a feature flag back on.

Sources: Anthropic Redeploying Fable 5 at https://www.anthropic.com/news/redeploying-fable-5; Platform release notes July 1, 2026 at https://platform.claude.com/docs/en/release-notes/overview.