Enable Automatic Model Fallbacks to Handle Safety Refusals Efficiently
Avoid failed requests by enabling server-side model redirection for blocked content.
Enable Anthropic's 'default' fallback mode in your API requests to automatically redirect safety-refused queries to a recommended alternative model.
The Scenario
You are building an application using Claude Fable 5, but users are occasionally getting refusals on technical topics due to conservative safety filters. You want to ensure they get an answer from a reliable model like Opus 4.8 without failing the request.
Before & after
Developers must manually write error-handling logic to catch "refusal" stop reasons and resubmit the same prompt to a different model, which can take 30–60 minutes to implement and debug.
Using the new 'default' fallback mode with the beta header, Claude automatically reroutes refused requests to a recommended model (like Opus 4.8) in milliseconds, requiring zero manual intervention.
The Prompt
// Use this configuration in your Anthropic API request to enable automatic fallbacks
{
"model": "claude-fable-5",
"fallbacks": ["default"],
"extra_headers": {
"beta": "server-side-fallback-2026-07-01"
},
"messages": [{"role": "user", "content": "[YOUR_PROMPT_HERE]"}]
}The 'fallbacks' parameter with the 'default' value allows Anthropic's server to handle model redirections when a request is declined by safety classifiers. This is particularly useful for Fable 5, which has conservative safeguards for biology and cybersecurity topics.
Source
Claude Platform release notes - Claude Platform Docs"set fallbacks to "default", and the API retries a declined request on the fallback model Anthropic recommends for its refusal category."