• 2 min read
Why pseudonyms may stop working online soon
A Dynomight essay argues that writing style alone may soon be enough to link pseudonymous accounts, even from small text samples.

Image: Hacker News
A new essay from Dynomight makes a stark claim: pseudonymous writing may become much harder to sustain, because text itself carries enough identifying information to link different identities back to the same person.
The piece, titled “Pseudpocalypse,” argues from a simple information-theory model. If every person had a unique binary signature, the tipping point for uniqueness in the Anglosphere would be about 29 bits, because 2^28.86 ≈ 490,000,000 — the number the author uses for currently alive Anglosphere residents. Reveal significantly fewer than 29 bits, and many people will still match you. Reveal significantly more, and the odds that someone else matches drop sharply.
Dynomight’s argument is that writing style may effectively leak those bits over time. Not through any literal ID string, but through recurring traits such as tone, word choice, punctuation, sentence structure, and stable linguistic preferences. The essay says an attacker would not need a database of every writer’s output to make this work; if two pseudonyms reveal enough overlapping features, the match itself could be enough.
The author says the idea feels increasingly plausible because authorship-guessing systems have improved quickly. Dynomight notes that after drafting most of the essay in mid-2025 and sitting on it for about a year, LLMs had already become much better at the task, adding that Claude 4.8 could identify the author from the first 1,000 words of a draft.
The article then broadens the idea beyond text. It suggests a more general “pseudpocalypse,” where people become identifiable through any sufficiently rich signal: body shape, gait, chemical signature, car sounds, tiny scratches, or even how someone moves a finger while scrolling.

Recommended reading
Schools urged to drop Google over AI Search risks
To ground the argument, Dynomight points to features that can plausibly be inferred from text:
- Demographics such as age, education, occupation, region, religion, sex, and native language
- Personality traits, including the HEXACO model’s six major dimensions and its 24 facets
- Stylistic markers such as word length, sentence length, punctuation use, function words, pronouns, modal verbs, hedges, conjunctions, and n-grams
The essay also places the idea in a longer history of stylometry. It cites Lorenzo Valla in 1440 on the Donation of Constantine, Augustus De Morgan in 1851 on average word length, and Mosteller and Wallace’s 1964 Bayesian analysis of The Federalist Papers, which concluded that Madison rather than Hamilton wrote all 12 of the disputed papers.
Dynomight’s bottom line is narrow but unsettling: if writing style contains at least 29 bits of usable information, and enough text is available, pseudonyms may not hold for long.
Security Editor
Sophia unpacks the invisible wars happening on our networks. Covering cybersecurity, privacy legislation, and cryptography, she exposes how our data is weaponized and defended. Before joining for(geeks), she spent years as a penetration tester. She's the reason the rest of the team uses physical security keys.
via Hacker News


