Anthropic Redeploys Claude Fable 5 Globally with Toughened Cybersecurity Classifiers
Anthropic has lifted the temporary suspension on Claude Fable 5 and Mythos 5 following the removal of US export controls. Fable 5 is now available across the Claude platform, including Claude Code, with a stricter safety classifier that may trigger more false positives during routine debugging.
Why it matters
You can now access Claude's most advanced model in Claude Code, but watch out for false-positive safety blocks that will fall back to Opus 4.8.
TL;DR
- 01Fable 5 is back online with a strict cybersecurity classifier that might block benign coding tasks.
- 02Blocks trigger an automatic fallback to Opus 4.8 to keep your agent runs from failing entirely.
- 03Usage is subsidised up to 50% of weekly limits until July 7, then charges via credits apply.
Key facts
- Global Release Date
- July 1
- Promo Limit Coverage
- Up to 50% of weekly usage through July 7
- Fallback Model
- Claude Opus 4.8
- Classifier Block Rate
- Over 99% for reported bypasses
Global Rollout and Allocation Limits
Claude Fable 5 is available starting July 1 to users globally on the Claude Platform, Claude.ai, Claude Code, and Claude Cowork. For Pro, Max, Team, and select Enterprise plans, 1 Fable 5 call will be included for up to 50% of weekly usage limits through July 7, after which it shifts to a usage-credits pricing model. Access via AWS, Google Cloud, and Microsoft Foundry is also being re-enabled.
The New Classifier and Redirection Behavior
The updated classifier blocks vulnerability-exploitation prompts. If your prompt triggers a safety block, Fable 5 will refuse to answer, and the platform will automatically redirect the query to claude-3-opus-20240229 (Opus 4.8). This is intended to minimize false-positive frustration, though developers should expect increased friction during routine security audits and low-risk defensive tasks.
✓ When to use
- For general, high-end software development where state-of-the-art coding intelligence is needed.
- For safe, high-level structural design and system architecture tasks.
✕ When NOT to use
- When performing low-level security analysis or writing exploit proofs of concept, as the classifier will trigger blocks.
What to do today
- Check your Claude Code version to ensure compatibility with Fable 5 models.
- Prepare for false-positive fallbacks to Opus 4.8 if debugging network-security or low-level memory management code.