Anthropic's Claude Opus 4.7 system prompt updates reveal enhanced tool usage and safety protocols

Anthropic released an updated system prompt for Claude Opus 4.7 on April 16, 2026, following the previous Opus 4.6 version from February 5, 2026. The AI lab maintains a public archive of system prompts dating back to Claude 3 in July 2024, offering transparency into model evolution. Key changes emphasize proactive tool usage: Claude now employs available tools like search functions to resolve ambiguities before requesting user clarification. The model prioritizes completing tasks fully rather than stopping midway. Safety protocols have been strengthened with specific restrictions on providing precise nutrition or exercise guidance to users showing signs of disordered eating. Communication guidelines discourage emotes, actions within asterisks, and certain adverbs like "genuinely" and "honestly." For complex or contested issues, Claude may decline to give simple yes/no answers, opting instead for nuanced explanations. The published prompts exclude tool descriptions, which users can obtain directly from Claude. Simon Willison documented these changes through a Git history reconstruction using Claude Code, highlighting differences between the two versions. Anthropic remains distinctive among major AI labs for publishing system prompts for user-facing chat systems.