Anthropic has updated Claude Opus 4.7's system prompt, marking the first change since February 2026. The update includes renaming the "developer platform" to Claude Platform and adding new tools like Claude in Powerpoint. The system prompt archive, dating back to Claude 3 in July 2024, shows how Anthropic refines their AI assistant's behavior over time.

Significant changes include expanded child safety instructions wrapped in a new critical tag, with explicit guidance that once Claude refuses a request for child safety reasons, subsequent conversation turns require extreme caution. The prompt now includes a new acting_vs_clarifying section that encourages the AI to make reasonable assumptions and use available tools before asking users for clarification. This reflects Anthropic's push toward more autonomous agent behavior.

The update removes language about avoiding certain words like "genuely" and "honestly," suggesting the new model has improved in this area. A new section addresses disordered eating concerns, prohibiting specific nutrition or exercise guidance. The system prompt also now guards against screenshot attacks by declining to give simple yes/no answers to complex or contested issues. These changes demonstrate Anthropic's ongoing refinement of Claude's safety boundaries and user interaction patterns.