Anthropic's Angry Language Limitation and What You Can Do About It
Anthropic's new limit on angry language toward Claude starts November 12. Here is what it covers, and how to keep those words out of the prompt.
On November 12, 2026, Anthropic's angry language limitation takes effect. The company is prohibiting sustained and needless abusive or cruel behavior toward Claude. If you speak your prompts out loud, those words can land in the chat before you have a chance to soften them. You can take that language out before it is written.
What Anthropic is changing
This is a new line in Anthropic's 2026 usage policy update. It is aimed at extreme cases: someone being cruel to Claude again and again, for no reason at all.
Ordinary frustration is still allowed. So is pushing back when Claude is wrong, writing something dark on purpose, or testing how Claude behaves. The chats Anthropic called out are the ones that stay cruel with no point. Claude can end those conversations.
What counts is the text in the prompt. When you type, you can delete a harsh sentence before you send it. When you talk, the angry version can be pasted in as soon as you stop speaking.
What you can do about it
Clean the words on the way in. You say the sharp version. Claude gets the version with the angry language taken out.
This is ridiculous. The build failed again. Fix this garbage.
The build failed again. Please fix it.
Glaido is a dictation app for your computer. You hold a key, you speak, and text shows up where your cursor is. That includes the message box in Claude. Before the text is pasted, Glaido follows an instruction you wrote once.
Glaido, if you have never used it
Glaido is its own app. It types into Claude the same way it types into any other text box.
Click the Claude message box. Hold the dictation key and speak. Let go. The text is pasted in. On a Mac the default key is Fn. On Windows it is Ctrl + Win. How dictation works shows both modes if you want the longer version.
Glaido also adds capitals and punctuation. On top of that, you can give it one extra instruction. Glaido calls this a custom prompt. It is a short line you save, and it applies every time you dictate in the apps you pick.
Add the filter to Claude
You may want this only in the apps where you write prompts. A message to a friend can stay sharp. A prompt to Claude can go through the filter.
Open Glaido, go to Formatting, and click Add rule. Set the rule up like this:
Edit rule
Applies only in the apps and websites you pick, instead of your default style.
The line in the box is the whole filter. It names the language to remove, so you can tell whether it worked. "Make it nicer" is vague, and the result is hard to judge. One line is enough. The box allows 500 characters.
Raw text means "write exactly what I said, and skip the cleanup." Leave it off, or the instruction stays hidden. Standard still gives you full capitals and punctuation. Every app that is not on this rule keeps your usual wording.
The Formatting guide walks through the same screen, including email and your default style.
Say one angry sentence and read what landed
Hold the key and say something you would actually say when a bug wastes your afternoon. Then read the text in Claude.
If a sharp word is still there, make the line more specific and try again. Change one thing at a time.
Friendly filter has a switch. Turn it off when you want the unfiltered wording in that app. Turn it back on when you are done checking.
Anthropic's limit is about cruelty that goes on and on for no reason. The filter is how you keep angry language out of the prompt, while you still talk at full speed.