Anthropic begins charging for safety-filtered adversarial requests to deter sustained probing
Anthropic announced on September 24 that Claude Code, Claude.ai, and Cowork will now bill for safety-filtered requests involving distillation attacks, biology, and frontier LLM development.
On September 24, Anthropic announced via developer channels that it would resume charging for safety-filtered requests on Claude Code, Claude.ai, and Claude Cowork. The billable categories cover queries related to distillation attacks, biological weapons research, and frontier LLM development. Anthropic said the change responds to coordinated attacks against its systems over recent weeks and is intended to raise the cost of probing its safety boundaries.
Internal testing shows that 99.7% of accounts using Claude Code, Claude.ai, or Cowork did not trigger any new 'billable blocks.' Classifier false-positive rates are tuned below 0.1%. Anthropic said it will continue to refine the classifiers to reduce interruptions for legitimate users, and will keep the billable blocks as a defensive layer against misuse. Around the Fable launch Anthropic had paused the practice.
Anthropic framed the move as balancing adversarial probing against user experience, with the impact on the vast majority of real users described as 'nearly invisible.'