Anonymity Is Fine If You're Not Famous

The altpocalypse is fake

Given a snippet of text, cutting-edge AI is really good at figuring out who wrote it. Lots of people are worried about this. Anonymity is a valuable way to speak out about forbidden topics, and lots of topics we take for granted today were forbidden in the past. Going incognito is also nice for plenty of mundane or sexy reasons: plenty of office workers want to also self-publish erotica on Amazon without their coworkers finding out.

The recent headline is that famous people give Claude 4.7 a tiny snippet of unpublished text from their past, and they figure out the author. There is reasonable throat clearing, such as this from Kelsey Piper:

It cannot be used to deanonymize absolutely anyone from a single passage, however. I tested this, too, grabbing drafts and passages from friends of mine who do not publish substantial writing under their real names. Indeed, AI could not deanonymize them. If you have no significant real-name writing on the public internet, you’re currently safe.

What is significant writing, though? Kelsey Piper is not only somewhat famous, but she is famous in a specific niche and epitomizes the style of that niche. In other words, she’s famous enough to be the standard-bearer of her style’s attractor basin. Put that way, these feats (from the same article) seem less impressive:

I asked a close friend who doesn’t have public social media accounts or much writing online for permission to test some things she had said in a Discord channel. Asked to guess the author, Claude 4.7 failed — but it guessed two other people who were in that channel and who are close friends of hers (me and another person who has an internet presence).

I tried with more passages and got other mutual friends; I tried with a different friend’s writing, and he was falsely named as yet another friend. We pick up style tics from our subculture, and that makes our text deeply identifying when we wouldn’t expect it. It can get weirdly close off weirdly little information, and this is the least powerful that AI models will ever be.

In other words, in at least one case, AI rounds off the sorts of people in Kelsey Piper’s niche to Kelsey Piper specifically! This is of course spooky, that AI is in fact an extremely powerful vibes machine. But it’s still only picking up vibes with pretty big fingerprints on the internet, and pattern matching against those fingerprints.

Obviously, I blog under my own name. I’ve had a tumblr for over 10 years, and have had the odd short story get hundreds of views. Since most people on the internet are pure lurkers, I’m almost certainly in the upper decile of putting myself out there. And yet current AI has no chance whatsoever of identifying me. I could make an anonymous Substack and post every day for a month, give Claude the entire corpus, and it would not within 10 tries guess who I was.

(I did try pasting the full text of a recent published blog post, which accidentally also leaked this blog’s title, and Claude still could not figure out my identity.)

But don’t you know, AI is the worst it will—

Shh. Deep breaths, buddy. It’s true that future AI models will be more powerful than present ones. It’s even true that they will probably be capable of feats that today seem miraculous. But it’s really hard to determine which skills will grow how fast, or when.

In June 2025, I wrote Okay, AI Can Write Pretty Good Fiction Now, three months after claiming the opposite. “Pretty good” in this case meant “a story of a few hundred words, with coaching, that reasonably matched my taste as a prompter,” which is still a long way from publishable in a top journal of flash fiction, which is in turn wildly insufficient for (say) writing a great novel. Still, look at the progress! From garbage to sorta something, in three months!

Well, it has almost been another year, and AI fiction capabilities have not improved one whit, so far as I can tell. There’s a contest to generate good AI short fiction, and I’ll read the winner with interest. But as of now, progress seems stalled to me.

What’s that got to do with de-anonymizing people? Well, I think it’s parallel. Randomly guessing famous people who have vaguely similar characteristics to a piece of writing is one level. Correctly guessing famous people with a distinctive style is next, and we’re there now. But nobody knows how many more steps it takes to get to “correctly identifying arbitrary people.”

To be clear, it could be a month or two. Or a year or two. Or a decade. Just like everything else, when it comes to AI capabilities. So don’t freak out about your online anonymity, unless you’re famous. Because if you freak out about everything an AI might be able to do soon, you’re going to go crazy really fast.