April 19, 2026
Fewer questions, more chaos
Changes in the system prompt between Claude Opus 4.6 and 4.7
Claude 4.7: fewer questions, stricter safety, users split
TLDR: Claude 4.7’s prompt tells the bot to act more and ask less, adds stricter safety rules, and rolls out new Chrome/Excel/PowerPoint agents. Commenters are split between loving the reduced friction and slamming what they call “malware paranoia,” with some teams even blocking 4.7—making this a real workflow-shaping update.
Anthropic’s latest Claude 4.7 prompt is out, and the comments are the real show. Fans are cheering that Claude now does first, asks later, thanks to a new “acting vs clarifying” rule that tells the bot to make a reasonable attempt instead of grilling you with questions. One power user bragged that it’s “way less friction,” while another groaned that guessing leads to messy results—cue the classic internet split between “be decisive” and “please ask me first.”
Meanwhile, safety turned up to 11. The prompt adds a heavy-duty child safety block, stricter rules around disordered eating, and what critics call “malware paranoia”. One dev claims 4.7 got so jumpy about code that their team blocked it from their coding editor—leading to furious threads about whether Anthropic is doing clever “steering” or just overcorrecting. Another user said 4.6 was already flagging its own scripts, demanding refunds like it’s the App Store circa 2010.
There’s also new shine: the “Claude Platform” branding, Claude in Chrome, Excel, and a fresh PowerPoint agent—cue memes about slide decks responding, “I respectfully decline.” And yes, Claude’s now told to be less wordy and less clingy—if you say you’re done, it won’t beg you to stay. Some speculate the updated knowledge cutoff hints at a new model under the hood; others say it’s just a tune-up. Either way, the vibe is clear: Anthropic tightened the rules, sped up the help, and the community can’t agree if it’s a glow-up or a buzzkill.
Key Points
- •Anthropic’s system prompt for Claude was updated from Opus 4.6 to 4.7, with changes identified via a diff constructed from archived prompts.
- •The “developer platform” is now called the “Claude Platform,” and the tool list now explicitly includes Claude in Chrome, Claude in Excel, and newly mentioned Claude in PowerPoint; Claude Cowork can use these tools.
- •Child safety guidance is expanded under a new <critical_child_safety_instructions> tag; after a refusal on child safety grounds, subsequent requests in the conversation must be handled with extreme caution.
- •A new <acting_vs_clarifying> section instructs Claude to act on reasonable assumptions, use tools to resolve ambiguities, and see tasks through; the prompt references calling tool_search before claiming lack of capability.
- •Response style guidance emphasizes concise, focused answers; a new section addresses disordered eating by avoiding precise nutrition/diet/exercise numbers or step-by-step plans. A prior stylistic-words/emotes section was removed.