Claude Code is scanning your messages for curse words | usagoldmines.com

“WTF?” “Dammit!” “Now I’m really annoyed.” Cursing out a flailing AI helper is something we’ve all done, but it turns out one of the most popular Claude tools is actively checking our messages for specific signs of frustration—including swear words.

The revelation is one of many startling details stemming from a massive Claude Code leak that laid bare many of Anthropic’s plans for its upcoming tools and models. The more than 500,000 lines of code, which Anthropic accidentally published to a public software registry on Tuesday, includes all kinds of juicy details, including specs for new Claude models, an “undercover mode” for Claude that allows it to make “stealth” contributions to public code bases, an “always-on” agent for Claude Code, and even a Tamagotchi-style “Buddy” for Claude.

But one of the stranger bits discovered in the leak is that Claude Code is actively watching our chat messages for words and phrases—including f-bombs and other curses—that serve as signs of user frustration.

Specifically, Claude Code includes a file called “userPromptKeywords.ts” with a simple pattern-matching tool called regex, which sweeps each and every message submitted to Claude for certain text matches. In this particular case, the regex pattern is watching for “wtf,” “wth,” “omfg,” “dumbass,” “horrible,” “awful,” “piece of —-” (insert your favorite four-letter word for that one), “f— you,” “screw this,” “this sucks,” and several other colorful metaphors. 

To be clear, this “swear word” search function was spotted specifically in Claude Code via the Anthropic leak. The code for Claude’s desktop and web apps wasn’t included in the leak, so we don’t know what’s going on under the hood of those Claude interfaces.

It’s also worth noting that the regex function used here is nothing fancy. Indeed, regex is supported by a wide variety of programming languages (everything from Java to Python) and has been in use for decades, and its functionality isn’t far removed from a plain-old Ctrl-F.

While the Claude Code leak revealed the existence of the “frustration words” regex, it doesn’t give any indication of why Claude Code is scouring messages for these words or what it’s doing with them.

I’ve reached out to Anthropic for comment.

Of course, we can guess. One plausible reason is that Anthropic is gathering telemetry to help it determine how specific Claude models and tools are performing. A boost in detected “frustration” words would be an easy way to spot if a new Claude feature isn’t going over well.

Another possibility is that a spike in detected frustration signals could trigger behavioral changes in Claude itself, perhaps making it more empathetic or apologetic. Of course, swearing at Claude or any other AI will generally steer the chat in a different direction (as it did for Google Search results), but a specific regex check like the one in Claude Code could help make a behavioral pivot more reliable.

While the “frustration words” regex has only been confirmed for Claude Code, it makes you wonder whether it exists in Claude’s desktop and web apps—or whether ChatGPT, Gemini, and other big AI players harbor similar functions in their code bases.