StegoStylo: Squelching Stylometric Scrutiny through Steganographic Stitching
Robert Dilworth · 2026
Read StegoStylo: Squelching Stylometric Scrutiny through Steganographic Stitching by Robert Dilworth (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-09056.txt.
Introduction
Robert Dilworth writes on how to hide who wrote a text from style checks. Style checks can name an author from word picks and pace. They aid law work and fraud hunts. They can also harm a source who must stay hid, or a writer in a strict land. Dilworth seeks a way to mask style while it keeps sense.
His method is called StegoStylo. It stitches text in a way that keeps the facts, but shifts the style marks. The aim is not to lie. It is to stop a style print from name the hand that wrote the words. A reader in text safety gets a mask plan, and a warn on how strong style prints now are.
Catalog record
| CREATOR | Robert Dilworth |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 29 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-09056
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
StegoStylo: Squelching Stylometric Scrutiny through Steganographic Stitching Robert Dilworth Abstract Stylometry--the identification of an author through analysis of a text’s style (i.e., authorship attribution)--serves many constructive purposes: it supports copyright and plagiarism investigations, aids detection of harmful content, offers exploratory cues for certain medical conditions (e.g., early signs of dementia or depression), provides historical context for literary works, and helps uncover misinformation and disinformation 1 1 1 The abstract mentions several benign use cases of stylometry. See the work of the following researchers as they relate to the positive implementation of stylometric analysis: plagiarism (academic-integrity investigations) - Albaqami et al. [ 4 ] ; harmful-content detection (filtering phishing emails) - Opara et al. [ 17 ] ; medical-condition discovery (predicting ADHD, depression, schizophrenia) - Barrios et al. [ 6 ] , Pérez et al. [ 19 ] , Rezaii et al. [ 20 ] ; providing historical context (composer identification through musical stylometry) - Abrosimov et al. [ 3 ] ; and misinformation/disinformation (detecting fake news) - Yang et al. [ 22 ] . . In contrast, when stylometry is employed as a tool for authorship verification–confirming whether a text truly originates from a claimed author–it can also be weaponized for malicious purposes. Techniques such as de-anonymization, re-identification, tracking, profiling, and downstream effects like censorship illustrate the privacy threats that stylometric analysis can enable. Building on these concerns, this paper further explores how adversarial stylometry combined with steganography can counteract stylometric analysis. We first present enhancements to our adversarial attack, TraceTarnish , providing stronger evidence of its capacity to confound stylometric systems and reduce their attribution and verification accuracy. Next, we examine how steganographic embedding can be fine-tuned to mask an author’s stylistic fingerprint, quantifying the level of authorship obfuscation achievable as a function of the proportion of words altered with zero-width Unicode characters. Based on our findings, steganographic coverage of 33% or higher seemingly ensures authorship obfuscation. Finally, we reflect on the ways stylometry can be used to undermine privacy and argue for the necessity of defensive tools like TraceTarnish . 1 New Research Angles Words can be like X-rays, if you use them properly–they’ll go through anything. You read and you’re pierced. Brave New World Aldous Huxley A key unanswered question from our preceding studies ( Dilworth [ 10 , 9 ] ) concerns the trade-off between anonymization quality and detectability. Specifically, how much must TraceTarnish –an adversarial stylometry attack that anonymizes text authorship–degrade a message to guarantee reliable anonymization? If perfect anonymization renders the text unreadable or nonsensical, its practical usefulness collapses regardless of resistance to detection. To address this, we must shift focus toward quantifying the optimal amount of “poisoning.” A promising approach is to measure the computed stylistic score of an original text alongside various steganographic embedding transformations, thereby identifying the degree of alteration that maximally obscures authorship while preserving readability. Inspired by Poisonify ( Jordan [ 13 ] )–an adversarial noise attack that sabotages AI music generation–and HarmonyCloak ( Meerza et al. [ 5 ] )–a defensive technique that shields music from generative models–we recognize untapped potential in how stego text (the result of embedding a hidden message within cover text) interacts with stylometric systems. In previous work, we sought to disrupt text-based pattern matching. For instance, if you wanted to search for a specific word within all the documents on your device, you could use the command-line utility grep 2 2 2 grep ’s manual page ( man page ) describes its functionality as a utility that “prints lines that match patterns.” to scour your files for a particular string or regular expression. All that is needed to thwart this search is to ensure that the utility never finds the pattern. One way to accomplish that is to inject a zero-width Unicode character as an infix 3 3 3 An infix, as defined by the Merriam-Webster dictionary, is “a[n]…affix appearing in the body of a word.” within the target string contained in a larger document. This is satisfactory; grep fails to detect the pattern, but a question arises: what is the extent of this embedding? As a disclaimer–and given the subject matter–it would be irresponsible not to provide a means of detecting the existence of invisible Unicode characters within the grep framework, as previously proposed. For those interested–or concerned–see the following resource [ 2 ] for guidance on how to discover such characters. For context, the stylometric application of grep involves using the tool to analyze textual patterns–such as word frequency, punctuation usage, and sentence structure–to assess an author’s unique writing style. Precomputed word lists, like the most common function words, are often used in stylometric analysis to identify and quantify specific vocabulary choices, allowing an investigator to detect stylistic similarities and differences between texts or authors. From a privacy standpoint, the intuition is that corrupting a word in a text that also appears in a word list–so that a search of the list for the corrupted word fails–undermines the grep -based stylometric scheme described earlier (and similar approaches). A question that emerges from this line of thinking is whether the hidden text can inflict damage on a stylometric system beyond merely causing pattern-recognition misses. Despite the tantalizing prospect, determining upper and lower bounds on the degree of encoding takes precedence; establishing these limits is essential before considering broader implications. The significance of this question is made more acute given the ramifications of stylometry. George Mikros, in their work ( Mikros [ 16 ] ), emphasizes that stylometry has valuable because “language provides insights about its creator,” and “every individual possesses a fairly consistent and unique linguistic ‘fingerprint’ (idolect) that can be observed in both spoken and written expressions.” As in our earlier work, our research aims to facilitate text-based interactions while disclosing as little information as possible about the participants–whether one-to-one (direct messages), one-to-many (publications, social media posts), etc.–and rendering each linguistic fingerprint less recognizable. This could, hypothetically, be achieved by processing all outgoing and incoming text with a tool like TraceTarnish , corrupting each participant’s stylistic fingerprint. In lieu of a conventional breakdown of our paper’s structure, we present a visual outline that better resonates with the essence of our content. As we delve into our work, the section headings we choose may divulge aspects of our identity best kept anonymous. Thus, we embrace the perfect opportunity to employ TraceTarnish . To illuminate the intricacies of this attack, a gentle suspension of disbelief may be required. Might every facet of this paper, save for the figure’s material, weave together a stylistic profile of its authors? The answer is a resounding yes. Yet, can you be certain that the text has not been poisoned with steganographic payloads meant to obscure such profiling? The answer rests in the shadows of debate. Regardless, we shall seize the opportunity nonetheless. See ( Figure 1 ) for an overview of our paper’s structure, where we showcase the various sections. However, the text has been corrupted using TraceTarnish . Figure 1: The structure of our paper, wherein the text has been compromised through the application of TraceTarnish . 2 Improvements to Our Existing Modularized Attack Script, TraceTarnish While we are relatively content with the state of our