Anthropic May Reroute 'Higher Risk' Requests to Sonnet 5

In one line: A report says that even when users pick Claude Sonnet 5.5, Anthropic may hand requests it classifies as "higher risk" to the earlier Sonnet 5 model.

Key points

• Per The New Stack, the model a user selects may not always be the model that processes the request.
• The trigger is reportedly requests flagged as "higher risk," apparently as part of a safety policy.
• The tension is that internal routing can intervene even when developers explicitly name a model via the API.

Why it matters

Model selection underpins output quality, cost, and reproducibility. If the chosen model can differ from the one that actually responds, it could affect result consistency in production pipelines and the reliability of benchmarks. Transparency and advance disclosure are likely to become the central debate.

Read more

• You picked Claude Sonnet 5.5 — but Anthropic may send your request to Sonnet 5 — The New Stack