Anthropic reportedly resumes charging for safeguard blocks before Claude replies
TestingCatalog’s September 25, 2026, brief lists biology, distillation attacks and frontier LLM work as affected categories. It says 99.7% of Claude Code, Claude.ai and Cowork accounts did not hit the new billable blocks.
TLDR
TestingCatalog reports in its September 25, 2026, AI brief that Anthropic resumed charging for certain safeguard blocks triggered before Claude replies. The listed categories are biology, distillation attacks and frontier LLM work. TestingCatalog says 99.7% of Claude Code, Claude.ai and Cowork accounts did not encounter the new billable blocks, and that classifiers were tuned to under 0.1% false positives. The publisher disclosed that it used Grok to compose the brief, selected the news and did some post-editing.
