FugitiveAI Deep Research — Knowledge Architecture and Content Depth

Full report

On this page

Executive Summary

FugitiveAI occupies a unique position within the educational technology landscape, utilizing self-contained, interactive puzzles to introduce fundamental concepts of hidden communication, structured data, and the representational divide between machines and humans. However, the current iteration relies on rudimentary modification-based steganography, lacking the comprehensive theoretical frameworks necessary to elevate the platform from a recreational puzzle site to an authoritative educational resource. This research report provides a blueprint for dramatically expanding FugitiveAI’s educational footprint. The analysis proposes formalizing a proprietary pedagogical framework termed "Machine Tradecraft," synthesizing established academic domains such as computational linguistics, information theory, digital forensics, and web accessibility standards. This framework will safely teach defensive analysis and structural literacy without exposing users to offensive security tools or unrestricted attack platforms. Critical expansion opportunities identified include the exploration of machine-readable metadata, Unicode manipulation (e.g., zero-width characters and homoglyphs), and the exploitation of web presentation layers, particularly the decoupling of Document Object Model (DOM) source order from visual rendering order. By integrating principles from digital forensics and information theory, the platform can impart a disciplined, scientific mindset. Furthermore, aligning this curriculum with local educational ecosystems—such as the cybersecurity and computer science programs at Morton College in Cicero, Illinois, alongside advanced research vectors from the University of Illinois Chicago (UIC) and DePaul University—positions FugitiveAI as a vital, self-contained preparatory environment for students entering the cybersecurity workforce.

1. Research Method

The investigation supporting this report relies on a comprehensive synthesis of primary and authoritative secondary sources across multiple disciplines. To establish the foundational knowledge architecture, the review targeted academic databases and repositories (ACM, IEEE, Springer, arXiv) for literature concerning linguistic steganography, digital watermarking, steganalysis, and covert network channels1. Particular attention was directed toward the evolution of modification-based steganography into modern generative steganography powered by large language models (LLMs)2. To address the differential interpretation of data by humans and machines, the research incorporated established web standards. Specifically, the analysis reviewed the Web Content Accessibility Guidelines (WCAG) managed by the W3C, focusing on the divergence between DOM source order and visual presentation via CSS manipulation5. Cybersecurity and digital forensics principles were drawn from academic research originating from regional institutions, including UIC and DePaul University8, alongside curriculum structures from Morton College in Cicero, Illinois, to anchor the pedagogical recommendations in a localized workforce-development context11. The evaluation strictly excluded literature proposing offensive exploitation, unrestricted prompt injection, or arbitrary third-party targeting. Terminology was carefully curated to distinguish between universally recognized academic concepts (e.g., lexical substitution, entropy, channel capacity) and proprietary pedagogical constructs necessary for FugitiveAI (e.g., "Machine Tradecraft").

2. Current Intellectual Position of FugitiveAI

FugitiveAI currently operates at the intersection of recreational cryptography, basic modification-based linguistic steganography, and entry-level metadata inspection. Its existing challenge families—such as acrostics, sentence-length parity, and synonym-choice encoding—utilize well-established historical techniques that rely on deterministic extraction rules13. However, evaluated against the broader academic consensus, FugitiveAI’s current theoretical scope remains shallow. In the formal study of linguistic steganography, techniques are broadly classified into modification-based methods (altering existing text without changing its semantic meaning) and generative methods (creating text that intrinsically contains the payload from inception)1. FugitiveAI currently teaches almost exclusively basic modification-based techniques. It has yet to explore advanced syntactic space hiding, such as manipulating dependency trees, or the profound complexities introduced by the LLM era3. Furthermore, the platform introduces machine-readable versus human-readable surfaces but lacks the robust vocabulary of digital forensics and information theory. By failing to formally introduce concepts such as entropy, channel capacity, normalization, and paraphrase destruction, the site currently treats each puzzle as an isolated novelty rather than a manifestation of a universal communication theorem. Moving forward, the platform must bridge the gap between abstract puzzles and real-world structural data concepts, accommodating structured metadata like JSON-LD16 and cross-script Unicode confusables17.

3. Machine Tradecraft Framework

"Machine Tradecraft" is not an established academic term within computer science, cryptography, or digital forensics. Therefore, it must be explicitly defined, internally contextualized, and presented as FugitiveAI’s proprietary pedagogical framework—an umbrella concept designed to teach how digital systems represent, transmit, and obscure information. To provide genuine educational value and avoid fabricating academic consensus, the Machine Tradecraft framework must organize established subjects rigorously. The proposed conceptual model operates along the following interacting dimensions:

• Carrier (The Medium): The seemingly innocuous artifact hosting the hidden data. This could be raw text, a DOM structure, semantic HTML, or Unicode strings.
• Representation Layer (The Viewport): The specific layer of interpretation. Information is processed differently depending on the observer. A visual rendering in a browser differs fundamentally from DOM tree inspection, which differs again from screen-reader audio output6.
• Transformation Rule (The Encoding): The deterministic logic applied to the secret data to map it into the carrier (e.g., converting bits to word-length parity choices).
• Covert Channel (The Vector): The specific structural, linguistic, or metadata anomaly exploited to transmit the payload without triggering normal human suspicion or automated filtering19.
• Channel Capacity & Robustness (The Limits): The mathematical and practical limits of the channel. This defines how much data can be hidden (capacity) versus how easily the data is destroyed by natural operations like copy-pasting or format conversion (robustness)20.
• Normalization Sensitivity (The Fragility): The degree to which extraction is impacted when the carrier undergoes standard data normalization, such as Unicode composed versus decomposed forms or whitespace stripping17.
• Verification & Extraction (The Proof): The deterministic, reproducible process a defender or analyst utilizes to isolate the signal from the noise, which is fundamental to the digital forensics mindset8.

By adopting this framework, FugitiveAI transitions from a repository of isolated puzzles to a systematic laboratory for analyzing any digital artifact.

4. Complete Knowledge Map

The following hierarchical taxonomy maps the expanded scope of FugitiveAI, moving systematically from foundational concepts to advanced inspection techniques.

This taxonomy serves as the architectural backbone of the site. Every puzzle, article, and glossary term will map directly to one of these nodes, ensuring that users understand where a specific technique fits within the broader landscape of digital information.

5. Major Subject Deep Dives

Classical Steganography

Classical steganography relies on hiding the existence of a message within a seemingly benign carrier. While historical methods like null ciphers and acrostics rely on human-driven linguistic tricks22, modern digital classical steganography heavily utilizes file formatting and metadata. Common techniques include Least Significant Bit (LSB) substitution in images or audio, where variations are imperceptible to human senses23. For FugitiveAI's harmless puzzle model, arbitrary file-upload analysis (like raw image hex editing) introduces excessive overhead and security risks. However, the conceptual foundation of LSB and whitespace manipulation can be safely taught through browser-based text and structural channels. The platform must explicitly distinguish steganography (hiding existence) from cryptography (hiding meaning) and watermarking (proving ownership or provenance)21.

Linguistic Steganography

Linguistic steganography is the concealment of information within natural language. Historically, it faced severe limitations regarding channel capacity and linguistic naturalness1. The discipline is divided into two primary vectors:

• Modification-based: Alters existing text through synonym substitution (lexical) or structural changes (syntactic)15. For example, replacing "make" with "create" based on a binary payload.
• Generative-based: The modern era utilizes LLMs to embed secret bits within the sampling randomness during text generation4. Techniques like Adaptive Dynamic Grouping (ADG) and Self-Adjusting Asymmetric Numeral Systems (SA-ANS) dynamically alter the token pool to maintain statistical indistinguishability while hiding data2.

FugitiveAI should lean heavily into modification-based linguistic steganography for its interactive puzzles, as these methods are deterministic and easily verifiable within a sandboxed browser. Generative steganography should be covered purely as theoretical advanced reading, educating users on its immense potential but warning them of its extreme fragility against paraphrase destruction and tokenization inconsistency4.

Machine-Visible versus Human-Visible Representation

A cornerstone of Machine Tradecraft is the realization that machines and humans parse data differently. The most profound educational opportunity within FugitiveAI lies in web accessibility and presentation layers.

• Source Order vs. Visual Order: With CSS Flexbox and Grid, developers can completely decouple the visual rendering of a page from the underlying HTML DOM order6. While a sighted user might read a page left-to-right, a machine parser or screen reader evaluates the DOM sequentially18. Hiding messages in the DOM order that are visually scrambled via CSS order or flex-direction: row-reverse is a powerful, harmless puzzle mechanic that simultaneously teaches WCAG accessibility standards (specifically Success Criterion 1.3.2: Meaningful Sequence)7.
• Invisible Metadata: Structured data formats like JSON-LD, microdata, and schema.org markup are consumed by search engines but ignored by casual human readers16. Steganographic payloads can be trivially tucked into these layers, operating as a sub-rosa channel that most users never inspect26.
• Unicode Manipulation: The complexity of the Unicode standard permits zero-width spaces (U+200B), zero-width non-joiners (U+200C), and homoglyph substitutions (visually identical characters from different scripts, such as a Cyrillic 'а' and a Latin 'a')17. These characters can bypass safety filters in AI systems27 and serve as incredibly rich, browser-safe puzzle vectors.

Information Theory

To elevate the curriculum, FugitiveAI must introduce foundational information theory. Puzzles should explicitly prompt users to consider the channel capacity of a given carrier—how many bits a paragraph can hold using word-parity versus synonym choice. Explanations must explore entropy (the measure of unpredictability) and redundancy (the natural padding in human language that steganography exploits). A critical lesson is the tension between robustness and detectability: a highly robust channel (one that survives copy-pasting across applications) is often highly detectable via statistical anomalies21.

Detection, Steganalysis, and Forensics

Defensive analysis requires a digital forensics mindset. Students at institutions like Morton College12 or UIC8 learn that forensics requires preserving the original artifact and documenting transformations. FugitiveAI can teach this by forcing users to carefully extract data without destroying fragile channels (e.g., losing zero-width characters during a careless copy-paste operation). Detection relies on finding statistical anomalies—for instance, if a text utilizes the passive voice 80% of the time in a language where 15% is the statistical norm, parser-assisted inspection will flag it as structurally unnatural.

6. Missing Foundations

A comprehensive audit of the established puzzle concepts reveals that FugitiveAI is currently missing several vital foundational prerequisites required to support deeper learning:

1. Text Normalization & Serialization: Users must understand how machines standardize text before processing it. Without understanding Unicode equivalence (NFC vs. NFD), bidirectional formatting, or whitespace stripping, users will struggle to understand why certain hidden channels break upon transmission across the internet17.

2. The Concept of Determinism: Puzzles rely on deterministic extraction (the rule always yields the exact same result). FugitiveAI must contrast this explicitly with probabilistic LLM generation to clear up recurring user confusion regarding how modern AI functions versus how classical encoding functions.

3. The Digital Forensics Mindset: The site currently treats puzzle-solving as a recreational game. It must reframe solving as forensic analysis: observing an artifact, formulating a hypothesis about the hidden channel, isolating the presentation layer, and applying a verifiable extraction rule8.

7. Existing Topic Expansion Opportunities

FugitiveAI’s existing challenges can be transformed into significantly deeper, authoritative lessons by connecting them to academic frameworks:

• Synonym-Choice Encoding: Currently taught as a simple word-swap mechanic. Expansion should introduce the concept of "Lexical Substitution Space," analyzing how semantic equivalence degrades if the synonym alters the context28. Furthermore, it should introduce statistical steganalysis: does the chosen synonym match the expected frequency distribution of the author's broader stylometry?29.
• Active/Passive Voice Choice: Currently framed as a basic grammar puzzle. Expansion should reframe this as "Syntactic Space Hiding"15. The curriculum can explain how dependency trees in computational linguistics map sentence structures, and how altering these trees carries a payload while maintaining perfect semantic coherence, offering higher security than mere lexical swaps.
• Punctuation & Spelling Variants: Expand into lessons on "Fragility and Normalization." Explain how spelling variants (e.g., British colour vs. American color) are highly robust against plaintext extraction, whereas punctuation channels are easily destroyed by automated grammar checkers, summarization tools, or strict serialization routines.
• Sentence-Order Encoding: Expand into the conflict between logical reading flow and steganographic payload. Use this to introduce the concept of "Paraphrase Destruction"—how quickly a sentence-order channel is destroyed if a human or AI rephrases the text to improve flow.

8. Cornerstone Content Plan

Cornerstone resources are comprehensive, 2,000+ word authoritative guides that serve as the foundation for the site's academic credibility and search authority.

9. Supporting Content Plan

Supporting articles (800–1,200 words) link out from cornerstones to provide specific, highly technical details on niche concepts without overwhelming the beginner.

1. Zero-Width Joiners and Non-Joiners Explained: Detailing how formatting characters act as invisible encoders30.

2. Homoglyph Attacks: Visual Spoofing in Plain Text: How cross-script confusables exploit human visual cognition17.

3. DOM Order vs. Visual Order: CSS Flexbox Disconnects: Technical breakdown of how order and flex-direction scramble layouts6.

4. Structured Metadata: Hiding Data in JSON-LD: Exploiting machine-readable sub-rosa channels16.

5. Least Significant Bit (LSB) Concept Simplified: Explaining the historical roots of digital steganography23.

6. LLM Tokenization Inconsistency in Steganography: Why generative steganography fails across different AI models4.

7. Paraphrase Destruction in Covert Channels: The impact of human rewriting on structured payloads.

8. Understanding Channel Capacity Limits in Text: Calculating bits-per-word ratios.

9. The Role of Redundancy in Language: How natural language padding creates the space for steganography.

10. Semantic Coherence in Lexical Substitution: Ensuring synonyms fit the surrounding context.

11. Statistical Frequency Analysis for Text: Detecting unnatural word choices.

12. HTML Comments and Deprecated Tags: Legacy hiding spaces in modern web development.

13. Whitespace and Typography Covert Channels: Exploiting trailing spaces and line breaks.

14. Steganography vs. Cryptography vs. Watermarking: Distinguishing between hiding existence, hiding meaning, and proving ownership21.

15. Base-4 and Base-64 Encoding Basics: Formatting data for transmission.

16. The Impact of Unicode Normalization (NFC/NFD): Why visually identical text has different byte values17.

17. ARIA Labels and Screen Reader Linearization: How accessibility trees process DOM nodes25.

18. Adaptive Dynamic Grouping (ADG) in AI Text: Modern techniques for dynamic token pooling2.

19. Hypothesis Validation in Forensics: The scientific method applied to artifact inspection.

20. Deterministic vs. Probabilistic Extraction: Why puzzles rely on rules, while AI relies on models.

10. Glossary Expansion

FugitiveAI requires a robust glossary to standardize terminology across all articles and puzzles.

1. Artifact: The digital object (text, code, image) being analyzed.

2. Capacity (Payload Capacity): The maximum amount of hidden data a carrier can hold without breaking structural or linguistic constraints20.

3. Carrier / Covertext: The seemingly innocent text or file housing the secret data.

4. Channel (Covert Channel): An unintended communication path used to transfer information in a way that violates expected systemic norms30.

5. DOM (Document Object Model): The machine-readable structure of a web page, which may differ from the visual render.

6. Entropy: A mathematical measure of the randomness or unpredictability of data within an artifact.

7. Extraction Rule: The deterministic logic required to decode a steganographic payload.

8. Homoglyph: Characters from different scripts that are visually identical or highly similar17.

9. JSON-LD: A machine-readable data format used for web metadata, commonly exploited for hidden data16.

10. Lexical Substitution: Hiding data by swapping words with their direct synonyms28.

11. Machine Tradecraft: FugitiveAI’s proprietary pedagogical framework for understanding digital representation and covert data.

12. Normalization: The process of standardizing data (e.g., removing extra spaces, resolving Unicode equivalences)17.

13. Paraphrase Destruction: The total loss of steganographic data when text is rewritten or summarized by a human or AI.

14. Representation Layer: The specific viewpoint (visual, audio, code, structural) from which data is observed.

15. Robustness: The ability of a hidden message to survive carrier modification or format shifting21.

16. Steganalysis: The analytical practice of detecting hidden messages24.

17. Stegosystem: The complete ecosystem comprising the encoder, decoder, payload, and carrier31.

18. Syntactic Hiding: Hiding data by altering the grammatical dependency structure of sentences15.

19. WCAG: Web Content Accessibility Guidelines, which inadvertently highlight how machines sequentialize web content32.

20. Zero-Width Character: Non-printing Unicode characters intended for formatting, often utilized maliciously to hide dense data streams17.

11. Comparison and Reference Content

Reference pages act as permanent utilities, drawing recurring traffic from learners and professionals alike.

1. Unicode Visually Confusable Characters Matrix: A table mapping common Latin, Cyrillic, and Greek homoglyphs.

2. Robustness vs. Capacity Trade-off Table: A matrix comparing techniques (e.g., whitespace vs. synonym choice) across survivability and data density.

3. DOM vs. Visual Order Checklist: A guide outlining how specific CSS properties (Flexbox, Grid, absolute positioning) decouple source order from visual rendering25.

4. Steganography Classification Tree: A visual flowchart separating modification-based from generative techniques.

5. Digital Forensics Order of Volatility (Text Version): A guide detailing what data gets destroyed first during copy-paste operations across operating systems.

6. Linguistic Features Extraction Matrix: Reference for word length, sentence length, vowel counts, and punctuation sets.

7. Common Metadata Surfaces: A checklist of HTML, MIME, and JSON-LD surfaces.

8. List of Zero-Width Unicode Points: Documentation of U+200B, U+200C, U+200D, etc., alongside their intended standard uses17.

9. Detection Vectors Field Guide: A step-by-step methodology for identifying statistical, visual, and structural anomalies in plain text.

10. Machine Tradecraft Taxonomy Poster: A downloadable, structured visualization of the core framework for classroom use.

12. Challenge-to-Education Architecture

To ensure users transition from merely playing a game to genuinely internalizing complex concepts, the pedagogy surrounding every puzzle must follow a strict, repeatable lifecycle.

1. Before Solving (The Hypothesis): The user is presented with a pristine artifact. They are taught to use a "Forensics Inspection Panel" (a strictly localized, browser-based tool) to view the artifact through different representation layers (e.g., Raw Text, DOM Tree, Visual Render).

2. Hints (The Pivot): Hints must never explicitly reveal the answer. They should direct the user to shift their representation layer or apply a specific normalization technique (e.g., "Have you tried looking at how a screen reader would parse this sequence?").

3. Solve (The Proof): The user applies the deterministic extraction rule to isolate the payload, reinforcing the scientific method.

4. Immediate Visual Explanation (The Revelation): Upon success, the UI dynamically highlights the artifact, visually revealing the covert channel in context (e.g., drawing connecting lines between parity words or highlighting zero-width injections).

5. Technique Article (The Local Context): A brief, embedded explanation of the specific mechanism (e.g., syntactic hiding).

6. Deeper Theory (The Global Context): Links to cornerstone articles (e.g., Information Theory, Robustness) to contextualize the puzzle.

7. Defensive Analysis (The Real-World Application): A concluding section explaining how a digital forensics analyst or automated steganalysis tool would detect this specific channel in the wild using statistical anomaly detection.

13. Visualization and Interactive Explanation Opportunities

Because FugitiveAI strictly prohibits unrestricted uploads, remote code execution, and arbitrary URL fetching, all interactive elements must be sandboxed, client-side JavaScript tools.

• The Representation Toggle: A core UI element allowing the user to view any puzzle artifact as Visual Render, Source HTML, DOM Tree, and Hexadecimal.
• Zero-Width Illuminator: An interactive text box that replaces invisible Unicode characters with bright, visible colored blocks, instantly revealing steganographic payloads and teaching character encodings29.
• CSS Order Scrambler: An interactive demonstration showing how changing CSS Flexbox order or flex-direction changes the visual layout instantly, while the screen-reader output (DOM order) remains completely static6.
• Synonym Tree Graph: A visual network showing a word and its synonyms, illustrating how choosing a lower-frequency synonym increases the channel capacity but simultaneously spikes the statistical detectability.
• Entropy Heatmap: A visual overlay on a paragraph of text where structurally unnatural or mathematically redundant sentences glow red, demonstrating how steganalysis detects anomalies.

14. Content Opportunity Matrix

The following matrix prioritizes the content build-out for FugitiveAI, balancing educational value, uniqueness, and implementation priority.

15. Research-Backed Prioritization

The roadmap prioritization is dictated by a strategic balance of educational value, implementation safety, and foundational necessity.

1. High Priority \- Representation Layers & Structural Literacy: Teaching users how to view HTML source, understand CSS decoupling, and inspect metadata is the highest leverage action. It bridges web development (WCAG accessibility) with security analysis, requiring zero backend infrastructure. It provides immense value to local institutions, such as Morton College's Computer Information Systems tracks, by reinforcing web fundamentals before students tackle advanced certifications like CompTIA Security+12.

2. High Priority \- Formalizing the Forensics Mindset: The platform already possesses functional puzzles. It immediately requires the analytical framework (Observation, Normalization, Verification) to contextualize those puzzles, preparing users for the rigors of programs akin to DePaul University’s Database Forensics curriculum10.

3. Medium Priority \- Information Theory: Introducing capacity, signal-to-noise ratios, and entropy requires some mathematical exposition but deeply legitimizes the platform's academic standing.

4. Low Priority \- LLM Generative Steganography: While heavily researched currently3, generative steganography relies on probabilistic models and external APIs, violating FugitiveAI's deterministic, self-contained puzzle ethos. It should only be covered theoretically in cornerstone articles.

16. Risks and Misconceptions

FugitiveAI must carefully navigate several pedagogical and reputational traps as it expands:

• Fabricating Academic Consensus: The term "Machine Tradecraft" must be clearly demarcated as a FugitiveAI educational tool, not an IEEE or ACM-recognized standard. Overstating its academic provenance will damage credibility.
• Conflating Cryptography with Steganography: Users frequently confuse encryption (making data mathematically unreadable) with steganography (hiding the mere existence of data). Content must sharply and repeatedly distinguish the two concepts2.
• Safety Filter Bypassing: Recent research highlights zero-width steganography as a severe threat vector for bypassing LLM safety guardrails (e.g., instructing a model via zero-width base-4 encoding)27. FugitiveAI must strictly frame zero-width characters as an artifact inspection concept and explicitly disclaim and refuse any utility that aids in constructing adversarial prompt injections.
• Oversimplification of Forensics: A browser-based tool cannot replicate a true forensics environment (e.g., memory carving or full disk imaging). The site must clarify that it teaches the foundational mindset of forensics (chain of observation, determinism), not the full technical discipline as taught in advanced academic centers8.

17. Recommended 90-Day Research/Content Roadmap

To deepen the existing product without requiring a structural redesign, the following sequence is recommended: Phase 1: Foundation and Taxonomy (Days 1–30)

• Draft and publish the "Introduction to Machine Tradecraft" manifesto.
• Restructure existing challenge categories to fit the new taxonomy (e.g., rename "Word-Length Puzzles" to "Lexical Modification Channels").
• Launch the comprehensive Glossary to standardize site-wide vocabulary.

Phase 2: The Representation Layer Deep Dive (Days 31–60)

• Develop and publish the cornerstone on DOM Order vs. Visual Order6.
• Implement the client-side "Representation Toggle" (Visual, Source, Hex) into the puzzle UI.
• Publish supporting articles on Web Accessibility (WCAG 1.3.2) and JSON-LD steganography.

Phase 3: Forensics and Analytics (Days 61–90)

• Publish the "Digital Forensics Mindset" cornerstone.
• Integrate post-solve "Defensive Analysis" breakdowns into all existing puzzles, explaining how statistical anomaly detection would flag the user's solved payload.
• Launch the Unicode and Homoglyph interactive reference matrices.

18. Top 10 Highest-Leverage Recommendations

1. Adopt the "Machine Tradecraft" Taxonomy: Transition the site from a loose collection of unrelated puzzles into a structured curriculum using the proposed framework.

2. Implement the Representation Toggle: Build a client-side tool allowing users to switch between visual rendering, DOM tree, and raw text on every puzzle artifact.

3. Bridge Security with Web Accessibility: Capitalize on the educational crossover between WCAG standards (logical reading flow) and CSS layout manipulation to teach DOM-level steganography7.

4. Publish the Digital Forensics Cornerstone: Reframe the act of puzzle-solving as formal artifact inspection, emphasizing artifact preservation and deterministic proof.

5. Build the Zero-Width Illuminator: Provide a safe, browser-based visualization tool that renders invisible Unicode characters visible, demystifying character encoding29.

6. Upgrade Synonym Puzzles to Lexical Steganography: Introduce the concepts of semantic equivalence, naturalness, and stylistic frequency analysis to existing simple puzzles.

7. Introduce Information Theory Basics: Add short, non-intimidating explanations of Channel Capacity and Robustness to the post-solve screens of existing challenges.

8. Create the Unicode Homoglyph Reference: Publish an exhaustive, interactive table of cross-script confusables17.

9. Delineate Modification vs. Generative Steganography: Explicitly outline why FugitiveAI uses modification-based techniques (for determinism) while providing theoretical overviews of how modern LLMs use generative steganography.

10. Target Local Educational Synergies: Align the curriculum progression to serve as a practical, interactive primer for students entering formal programs at institutions like Morton College (CompTIA Security+, CIS)12, explicitly bridging basic IT fundamentals with the forensic concepts studied at higher-level regional research institutions.

Sources

• \[cite: 20\] Linguistic Steganography via Self-Adjusting Asymmetric Numeral System (MIT Press / Computational Linguistics) \- Supports LLM-based generative steganography adaptations.
• \[cite: 23\] Comparison of Steganography Technology (IJETT) \- Supports the foundational definitions of LSB and spatial vs. frequency domain steganography.
• \[cite: 13\] Review of Text Steganography Methods (MDPI) \- Supports classification of text steganography into statistical, format-based, and linguistic methodologies.
• \[cite: 1\] A Comprehensive Survey on Linguistic Steganography (ACM / ResearchGate) \- Supports the historical constraints of linguistic steganography and the transition into the LLM era.
• \[cite: 22\] Linguistic Steganography: Survey, Analysis, and Robustness (Purdue University CERIAS) \- Supports the analysis of natural language cover texts and the necessity of semantic cohesiveness.
• \[cite: 2\] Adaptive Dynamic Grouping for Generative Linguistic Steganography (arXiv) \- Supports modern techniques for dynamic token grouping and embedding capacity in AI text.
• \[cite: 14\] Multi-Time-Step-Based Steganography Method (PMC) \- Supports the integration of multiple time steps to improve statistical indistinguishability in generated text.
• \[cite: 3\] A Comprehensive Survey on Linguistic Steganography: Methods, Countermeasures, Evaluation, and Challenges (OpenReview) \- Supports the taxonomy of countermeasures and evaluation metrics.
• \[cite: 28\] Linguistic Steganography for Messaging Applications (ResearchGate) \- Supports lexical substitution, synonymous swapping, and the maintenance of semantic equivalence.
• \[cite: 24\] Linguistic Steganography and Linguistic Steganalysis (ResearchGate) \- Supports defensive steganalysis techniques, multi-bit watermarking, and distinguishing steganography from watermarking.
• \[cite: 15\] Linguistic Steganography: Hiding Information in Syntax Space (IEEE) \- Supports syntactic hiding via sentence-level structural manipulation and dependency trees.
• \[cite: 1\] A Comprehensive Survey on Linguistic Steganography (Preprint) \- Supports definitions of modification-based versus generative steganographic methods.
• \[cite: 4\] Addressing Tokenization Inconsistency in Steganography (ResearchGate) \- Supports the analysis of fragility, tokenization errors, and segmentation ambiguity in modern LLM steganography.
• \[cite: 19\] A Discussion of Covert Channels and Steganography (Semantics Scholar) \- Supports the foundational definitions of covert channels in web environments.
• \[cite: 8\] Overview of Cyber-Resilient Smart Inverters (Forensics) (UIC Research) \- Supports the necessity of digital forensics, artifact extraction, and preservation methodologies.
• \[cite: 8\] Cybersecurity Threats and Forensics (UIC Research) \- Supports localized academic context regarding operational technology and real-time intrusion detection.
• \[cite: 9\] Social Forensics: Searching for Needles in Digital Haystacks (UIC Publications) \- Supports the concepts of systematic characterization and anomaly detection in digital data.
• \[cite: 10\] Database Forensic File Format and Toolkit (DePaul University) \- Supports the requirement of abstraction and standardization in forensic artifacts.
• \[cite: 11\] Cybersecurity Boot Camps (Morton College) \- Supports localized educational context and curriculum alignment for introductory cybersecurity students.
• \[cite: 33\] Information Technology Programs (Morton College) \- Supports the connection between foundational IT training (CompTIA) and FugitiveAI's target audience.
• \[cite: 12\] Computer Information Systems Curriculum (Morton College) \- Supports the integration of web design, multimedia, and network security as prerequisite knowledge bases.
• \[cite: 29\] Zero-Width Steganography and Stylometric Attribution (arXiv) \- Supports the use of zero-width characters as a covert channel and their impact on forensic analysis.
• \[cite: 27\] LLM Steganographic Covert Channel (Promptfoo / LLM Security Database) \- Supports the risks of zero-width Unicode injection bypassing AI safety filters.
• \[cite: 17\] Unicode Control Character Attacks (arXiv) \- Supports the vulnerability of systems to homoglyphs, zero-width joiners, and bidirectional overrides.
• \[cite: 31\] Agentic Covert-Channel Construction (arXiv) \- Supports the definition of a stegosystem (encoder, decoder, payload space) and focal-point coordination.
• \[cite: 21\] Dual Stage Text Steganography Using Unicode Homoglyphs (ResearchGate) \- Supports the practical application of space manipulation and homoglyphs in text steganography.
• \[cite: 17\] Systematic Classification of Character-Level Attacks (arXiv) \- Supports the taxonomy of Unicode manipulation, script confusion, and encoding obfuscation.
• \[cite: 21\] Content-Preserving Text Watermarking (ResearchGate) \- Supports the distinction between visual indistinguishability, channel capacity, and robustness.
• \[cite: 30\] Reversible Storage Network Covert Channels (ResearchGate) \- Supports the use of zero-width joiners/non-joiners to conceal data with high capacity.
• \[cite: 5\] WCAG 2.0 Techniques: Making DOM Order Match Visual Order (W3C) \- Supports the foundational accessibility standard requiring source order to match visual presentation.
• \[cite: 32\] EPUB Accessibility: Source Order vs. Visual Order (W3C) \- Supports the programmatic requirement of reading order in structured documents.
• \[cite: 18\] Source Order Matters (Adrian Roselli) \- Supports the technical analysis of CSS grids and flexbox disrupting the DOM sequence for assistive technologies.
• \[cite: 34\] MDN Web Docs: Reading Order (Mozilla) \- Supports the definition of reading flow and its impact on sequential navigation and screen readers.
• \[cite: 25\] WCAG Criterion 1.3.2: Meaningful Sequence (EqualWeb) \- Supports the compliance requirements for screen readers linearizing DOM order against CSS reordering.
• \[cite: 25\] How to Meet 1.3.2 (EqualWeb) \- Supports the identification of common failures in multi-column layouts and CSS Flexbox ordering.
• \[cite: 6\] Document Order, Visual Order, and Accessibility (Rivendellweb) \- Supports the classification of visual reordering risks and their human impact on cognitive accessibility.
• \[cite: 6\] Implementation Guidance for CSS Grid and Flexbox (Rivendellweb) \- Supports the technical execution of placing items without altering interaction sequences.
• \[cite: 7\] WCAG Technique C27 (W3C) \- Supports the objective of ensuring source code content order mirrors visual presentation to prevent assistive technology confusion.
• \[cite: 16\] Data Hidden in Plain Sight (Stephen Hawkes) \- Supports the practical application of tucking steganographic JSON-LD into reusable image components.
• \[cite: 26\] Steganography applied to Microdata (Semantics Scholar) \- Supports the integration of steganography with structured metadata to bypass similarity checks.

Works cited

1. A Comprehensive Survey on Linguistic Steganography: Methods, Countermeasures, Evaluation, and Challenges \- ResearchGate, https://www.researchgate.net/publication/398488225\_A\_Comprehensive\_Survey\_on\_Linguistic\_Steganography\_Methods\_Countermeasures\_Evaluation\_and\_Challenges

2. Provably Secure Generative Linguistic Steganography \- arXiv, https://arxiv.org/html/2106.02011v1

3. A Comprehensive Survey on Linguistic Steganography: Methods, Countermeasures, Evaluation, and Challenges | OpenReview, https://openreview.net/forum?id=sjgSCTaPCu

4. Ruiyi YAN | PhD Student | Kyoto University, Kyoto | Kyodai | Research profile \- ResearchGate, https://www.researchgate.net/profile/Ruiyi-Yan-5

5. C27: Making the DOM order match the visual order | Techniques for WCAG 2.0 \- W3C, https://www.w3.org/TR/2014/NOTE-WCAG20-TECHS-20140408/C27

6. Document order, visual order, and accessibility \- The Publishing Project, https://publishing-project.rivendellweb.net/document-order-visual-order-and-accessibility/

7. Technique C27 Making the DOM order match the visual order. \- W3C, https://www.w3.org/WAI/WCAG21/Techniques/css/C27

8. An Overview of Cyber-Resilient Smart Inverters Based on Practical Attack Models \- Laboratory for Energy and Switching-Electronics Systems \- University of Illinois Chicago, https://mazumder.lab.uic.edu/wp-content/uploads/sites/504/2024/02/T\_An\_Overview\_of\_Cyber-Resilient\_Smart\_Inverters\_Based\_on\_Practical\_Attack\_Models.pdf

9. Jason Polakis \- Computer Science, https://www.cs.uic.edu/\~polakis/polakis-cv.pdf

10. DB3F and DF-Toolkit, http://dbgroup.cdm.depaul.edu/DF-Toolkit.html

11. Cybersecurity Programs \- Morton College, https://careertraining.morton.edu/information-technology-cybersecurity-programs/

12. Computer Information Systems | Advance Your Tech Skills Today \- Morton College, https://www.morton.edu/cis

13. A Review on Text Steganography Techniques \- MDPI, https://www.mdpi.com/2227-7390/9/21/2829

14. MTS-Stega: Linguistic Steganography Based on Multi-Time-Step \- PMC, https://pmc.ncbi.nlm.nih.gov/articles/PMC9140853/

15. Linguistic Steganography: Hiding Information in Syntax Space | Request PDF \- ResearchGate, https://www.researchgate.net/publication/376840221\_Linguistic\_Steganography\_Hiding\_Information\_in\_Syntax\_Space

16. Data Hidden in Plain Sight \- Stephen Hawkes, https://stephenhawkes.com/blog/2025/05/06/data-hidden-in-plain-sight/

17. Special-Character Adversarial Attacks on Open-Source Language Models \- arXiv, https://arxiv.org/html/2508.14070v1

18. Source Order Matters \- Adrian Roselli, http://adrianroselli.com/2015/09/source-order-matters.html

19. \[PDF\] A Discussion of Covert Channels and Steganography, https://www.semanticscholar.org/paper/06b80f4280e842f93526b9e6f239a7de9c07c4ea

20. Linguistic Steganography via Self-Adjusting Asymmetric Number System \- MIT Press Direct, https://direct.mit.edu/coli/article/52/1/113/132854/Linguistic-Steganography-via-Self-Adjusting

21. Dual Stage Text Steganography Using Unicode Homoglyphs \- ResearchGate, https://www.researchgate.net/publication/300555812\_Dual\_Stage\_Text\_Steganography\_Using\_Unicode\_Homoglyphs

22. CERIAS Tech Report 2004-13 LINGUISTIC STEGANOGRAPHY: SURVEY, ANALYSIS, AND ROBUSTNESS CONCERNS FOR HIDING INFORMATION IN TEXT by, https://www.cerias.purdue.edu/tools\_and\_resources/bibtex\_archive/archive/2004-13.pdf

23. A Survey on various types of Steganography and Analysis of Hiding Techniques, https://ijettjournal.org/archive/ijett-v11p276

24. Linguistic Steganography and Linguistic Steganalysis | Request PDF \- ResearchGate, https://www.researchgate.net/publication/378720604\_Linguistic\_Steganography\_and\_Linguistic\_Steganalysis

25. WCAG 1.3.2 Meaningful Sequence \- How to Meet It \- EqualWeb, https://www.equalweb.com/wcag/criteria/1-3-2/

26. A novel privacy-preserving technique using steganography and L, https://www.semanticscholar.org/paper/1445f20b67b86c9d36c991aae4e7f99cf72dcd7b

27. LLM Steganographic Covert Channel | LLM Security Database \- Promptfoo, https://www.promptfoo.dev/lm-security-db/vuln/llm-steganographic-covert-channel-dd117e1d/

28. Linguistic Steganography for Messaging Applications | Request PDF \- ResearchGate, https://www.researchgate.net/publication/358630302\_Linguistic\_Steganography\_for\_Messaging\_Applications

29. Unveiling Unicode's Unseen Underpinnings in Undermining Authorship Attribution \- arXiv, https://arxiv.org/html/2508.15840v1

30. (PDF) Towards Reversible Storage Network Covert Channels \- ResearchGate, https://www.researchgate.net/publication/333996207\_Towards\_Reversible\_Storage\_Network\_Covert\_Channels

31. Tool Use Enables Undetectable Steganography in Multi-Agent LLM Systems \- arXiv, https://arxiv.org/html/2606.28425v1

32. EPUB Accessibility \- Fixed Layout Challenges and Best Practices \- W3C, https://www.w3.org/TR/epub-fxl-a11y/

33. Information Technology \- Morton College, https://careertraining.morton.edu/information-technology-programs/