Anthropic resumes billing for safety-blocked requests; 99.7% of users unaffected

ClaudeDevs · x · 2026-09-25

Anthropic says it will resume charging for requests its safeguards block before Claude responds, limited to low-false-positive categories (biology, distillation attacks, frontier LLM development), citing coordinated attacks. It reports 99.7% of users won't hit these billable blocks, classifiers are tuned to <0.1% false positives, and misfires can be reported via /feedback in Claude Code.

Original post →

More from Models

Models channel →