|
Happy Sunday,
Today I'm going back to the Claude watermark story because a lot more has come out since Wednesday and someone has now built a free tool to strip them out, so I'm showing you what it really does, then telling you about the American city that just handed some of its 911 calls over to an AI, and finishing with two prompts that pull apart what your last name actually means and where it came from.
Let's get into it.
|
IN THIS EMAIL
» Someone built a Claude watermark remover
» A city just gave its 911 calls to an AI
» What your last name actually means
|
🚨 THE FULL STORY
Someone Built a Tool to Remove Claude's Watermarks
On Wednesday I told you Claude had started hiding an invisible watermark in everything it writes, and I talked about some of the upsides of that. A lot more has come out since then, so let me take you through it properly, because most of what I've read on this in the last few days has been wrong.
The watermark is not a little tag stuck on the end that you could delete. It's hidden in the words themselves. When Claude writes, it very slightly nudges which words it picks, in a pattern a machine can recognise later. That's why it survives copy and paste, and why changing a few sentences doesn't shift it. There is nothing to find and delete. The pattern is the writing. And worth remembering, they're only doing any of this because Europe's new AI law makes them. They just decided to switch it on for the entire world instead of only Europe.
And here's the part that's really freaking people out. Say you write something yourself, your own idea, your own words, and you paste it into Claude to tidy up the grammar. What comes back is now marked as AI. Not the grammar fix. The whole thing. Your thinking, flagged as machine output because a machine touched the punctuation. Anthropic admit this themselves, they say finding the watermark doesn't prove Claude wrote it. So they know it doesn't mean what people are going to assume it means, and they rolled it out to the entire planet anyway. That's the part I don't love.
So somebody built a free tool to strip these marks back out. It's called Watermarks Remover, it's open source, and it's by a developer named Guillaume Meyer. Worth knowing it isn't only for Claude either, it handles the marks from ChatGPT and Gemini too. I went through it properly, so here's the honest version of what it does.
What works: AI tools leave little bits of invisible junk in your text, characters you can't see that sit there quietly announcing where it came from. This wipes them out completely, and does the same for the hidden details buried inside image and document files. That part is real and it works.
What we don't know yet: the watermark itself lives in the word choices, so the tool has a second job for that one. It goes in and rewrites your text to shuffle the patterns around. Whether that's enough to beat the watermark, nobody can tell you yet. Not the guy who built it, not me, not anyone, because there's no detector out there to test it against. So it might work perfectly. It might not. That's genuinely where we are, and it's worth installing now so it's sitting there ready either way.
Here's how to install it, and it takes about a minute.
Step 1: Open the project page, then click into the skills folder (it's the fourth one down), then into the file called SKILL.md, and choose Download raw file.
Step 2: Come over to Claude. On the left choose Customize, then Skills.
Step 3: Hit the dropdown in the top right, choose Upload skill, and drag the file in.
Step 4: Type /remove-ai-marks and paste your text in.
And this is the bit that should calm everybody down. There is no detector yet. Not for anyone. Anthropic say they're building the tools to read these marks but haven't released them, so no teacher, boss or client can check your writing for this today. The marks are going in, the ability to read them comes later. The moment they release it we'll finally be able to test whether any of these removers actually do the job, and I'll tell you either way. In the meantime nobody needs to panic, and the thing that always worked still works best. Use AI to think and to draft, then rewrite it properly in your own voice, and it's yours in every way that counts.
Get my full watermark setup guide →
📞 THIS IS WILD
A City Just Gave Its 911 Calls to an AI
New Orleans has started letting an AI answer some of its 911 calls. Read that back once, because I had to.
Now before anyone panics, it is not replacing the humans, and a lot of the coverage of this got that wrong. What it handles is the pile-up. When there's a crash on the highway at rush hour, forty people ring 911 about the same crash, and every one of those calls ties up somebody who could be talking to a person having a heart attack. The AI picks up the duplicates, the incident already logged and already being dealt with, and hands everything else straight to a human.
They trained it on real 911 recordings, including local street names and the way people there actually speak, which matters more than it sounds when someone is panicking and slurring an address. In its first 90 days it sorted over 3,500 calls, cut the duplicate ones by more than 30%, and pulled response times down by up to 40 seconds. In an emergency, 40 seconds is not a rounding error.
Here's the bit nobody's really saying out loud though. This didn't happen because someone got excited about AI. It happened because roughly a third of their call-taking jobs are empty and they can't fill them. Seattle and Atlanta are already doing versions of this. That's a US story, but a 911 centre that can't hire enough people is not a US problem, and every country reading this has the same shortage in its own emergency services.
And I go back and forth on it. Critics worry an AI mishears someone in real distress, or misses that a voice is terrified rather than annoyed, and that's fair. But the alternative on offer isn't a fully staffed room of humans. It's being on hold. Once you frame it that way I think it's the right call. Just watch it closely, because this is AI moving out of your laptop and into the number you ring on the worst day of your life.
Read the full story here →
🧠 TRY THIS TODAY
What Your Last Name Actually Means
This one takes about a minute and the input is one word. Your last name. You've written it ten thousand times and there's a decent chance you've never once asked where it came from or what it meant to the first person who carried it. Works in ChatGPT or Claude, either is fine.
Step 1: Paste this in.
|
My last name is [YOUR LAST NAME]. Tell me everything about it. What language it comes from, what it literally translated to when it was first used, and why a family would have been given that name in the first place, whether it was a job, a place, a father's name, or something about how someone looked. Tell me roughly when surnames like mine started being used, what my ancestors were most likely doing for a living, and where in the world the name is most common today. If there are competing theories about the origin, give me all of them and tell me which one is best supported and why.
|
Most surnames turn out to be one of four things. A job somebody did, a place they lived, whose son they were, or a nickname about their appearance that somehow got stuck onto every generation since. Somewhere back there a real person got handed that word, and everyone who has ever shared your name is downstream of them.
Step 2: Now the one that gets people. Stay in the same chat and say:
|
Now trace the journey. Where did this name start, how did it spread across the world, and why did families carrying it end up where they are now. Then tell me every way the spelling has changed over the centuries, and what my name was most likely spelled as before it became this, especially if it was changed or shortened when a family moved country. Tell me what you are confident about and what you are guessing at, and be honest about the difference.
|
That second one is the payoff, because an enormous number of surnames got changed at some point. Shortened because nobody could spell it, altered on arrival in a new country, or quietly adjusted by a family who wanted to blend in. So the name you've had your whole life might be a version of something else, and that's usually the moment this stops being a fun prompt and turns into a phone call to the oldest person in your family.
One honest caveat. It's working from patterns in how names formed, not from your actual family tree, so treat it as a strong starting point rather than proof. That last line asking it to separate what it knows from what it's guessing is doing real work, don't drop it. And this works wherever in the world your family is from, which isn't true of most things I send you.
Get my full family history guide →
Do the last name one today, it's a Sunday and it takes a minute. And if it turns up a spelling nobody in your family has mentioned, go and ask the oldest person you're related to about it before you lose the chance. That's a better use of an afternoon than anything else in this email.
See you Wednesday.
Maverick
P.S. Want your business to actually show up inside AI? The brands that rank are seeing up to 3x the conversions. Get your free AI visibility score and see exactly where you stand inside AI.
P.P.S. If you're new here or you ever skip one, I've got a newsletter archive of all my past emails you can catch up on anytime.
|
