Changes in the system prompt between Claude Opus 4.6 and 4.7

2026-04-18 · Source: Simon Willison's Weblog · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, medium

Summary

Anthropic updated the system prompt for Claude Opus 4.7, released on April 16, 2026, building upon the previous Opus 4.6 prompt from February 5, 2026. Key changes include renaming the "developer platform" to "Claude Platform" and expanding the list of integrated tools to include "Claude in Chrome", "Claude in Excel", and "Claude in Powerpoint". The child safety section was significantly enhanced with a new `<child_safety>` tag, emphasizing extreme caution after an initial refusal. New directives aim to make Claude less verbose and less pushy, respecting user requests to end conversations. The prompt also introduces a `tool_search` mechanism, encouraging Claude to actively look for tools to resolve ambiguities before asking the user. Specific language addressing "disordered eating" and a new guard against forced yes/no answers on complex issues were added. The political clarification regarding Donald Trump's presidency was removed, reflecting an updated knowledge cut-off date.

Key takeaway

For AI Engineers developing conversational agents, analyzing system prompt evolution, as demonstrated by Anthropic's Claude updates, is crucial. You should review your model's core directives to ensure they align with evolving safety standards, user experience expectations (e.g., less pushy, more concise), and advanced tool integration capabilities. Consider implementing a `tool_search` mechanism to empower your agents to resolve ambiguities autonomously, enhancing user interaction and reducing unnecessary clarification prompts.

Key insights

Anthropic's Claude Opus 4.7 system prompt updates enhance safety, user experience, and tool integration.

Principles

• Prioritize user autonomy and safety
• Act with tools before asking users
• Maintain concise, focused responses

Method

Anthropic uses a Git history approach to track system prompt evolution, allowing for detailed diff analysis between model versions like Opus 4.6 and 4.7 to identify specific changes.

In practice

• Integrate `tool_search` for dynamic capability discovery
• Implement specific safety tags like `<child_safety>`
• Refine prompt to reduce verbosity and pushiness

Topics

• Claude Opus 4.7
• System Prompt Updates
• AI Model Behavior
• Child Safety Guidelines
• Tool Integration

Code references

• simonw/research

Best for: AI Engineer, NLP Engineer, Prompt Engineer, AI Scientist, Machine Learning Engineer

Related on AIssential

• Anthropic’s Claude Opus 4.7 is finally here — Claude Opus 4.7 improves coding and reasoning, but new safety measures slightly reduce cybersecurity vulnerability reproduction.
• AI and Claude Design: Will it Replace Designers by 2026? — Claude Design applies Anthropic's AI, like Claude Opus 4.7, to design workflows, sparking debate on AI's role in design.
• llm-anthropic 0.25 — The `llm-anthropic` library update adds new Claude models and enhanced control over model "thinking" processes.
• v2.1.113 — Recent Claude Code updates enhance security, performance, and agent capabilities through new features and bug fixes.
• I Watched the Entire Anthropic Workshop and Here Is a Recap — Engineers often underutilize agentic coding tools, treating them as autocomplete instead of autonomous execution engines.
• Claude Opus 4.6 First Impressions & More AI News You Can Use — Claude Opus 4.6 significantly advances agentic AI capabilities, while OpenAI targets enterprise automation with custom agents.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Simon Willison's Weblog.