GHS for Dark Patterns

• GHS for Dark Patterns is a structured framework that classifies, labels, and regulates deceptive UI/UX designs to protect user autonomy.
• It integrates diverse taxonomies through directed graph analysis, hierarchical ontologies, and community detection methods to consolidate dark pattern typologies.
• The system employs standardized glyphs and formal decision rules to enable automated detection, audit, and cross-jurisdictional regulatory enforcement.

A Globally Harmonized System (GHS) for Dark Patterns constitutes a structured approach to classifying, labeling, and regulating manipulative UI/UX design practices that subvert user autonomy and distort choice architecture. The drive for harmonization arises due to the proliferation of overlapping, domain-specific taxonomies and inconsistent terminologies, which have impeded the emergence of enforceable global standards and the interoperability of detection, auditing, and legal redress workflows. Unifying research contributions from the fields of HCI, legal studies, and regulatory science, a GHS aims to formalize dark pattern typologies, define standardized labeling schemas (including glyph-based warnings), and supply rigorous methodologies for extension, empirical validation, and translational governance (Lewis et al., 2024, Gray et al., 2023, Brenncke, 2023).

1. Origins and Motivations for Harmonization

The identification and mitigation of dark patterns has historically been fragmented, with at least sixteen influential taxonomies mapping out heterogeneous pattern sets across domains such as e-commerce, privacy, gaming, and data protection. Notable examples include Brignull’s (2010) taxonomy of deceptive UI tricks, Greenberg et al.’s (2014) analysis of proxemic dark patterns, Mathur et al.’s (2019) empirical crawl of shopping UIs, and Bosch et al.’s (2016) antithesis of privacy-by-design. Each framework offers distinct, sometimes overlapping, pattern definitions—ranging from “Confirmshaming” and “Bait & Switch” to “Information Hiding,” “Nagging,” and “Maximize” (Lewis et al., 2024). The lack of interoperability and convergent lexicon across these schemes limits their practical utility for global regulation.

Regulatory momentum, particularly in the European Union, frames the mitigation of dark patterns as an autonomy-preservation challenge, with legal statutes increasingly referencing behavioral manipulations such as drip pricing, misleading salience, and friction engineering (Brenncke, 2023). The harmonization imperative thus emerges to unify research, regulatory, and practitioner vocabularies—enabling systematic detection, enforcement, and cross-jurisdictional translation.

2. Typological Frameworks: Three Principal Models

Three major harmonization proposals shape the GHS discourse:

1. Directed-Graph Integration and Cluster Detection (Lewis et al., 2024): Synthesizes sixteen taxonomies into a directed, unweighted graph $G = (V, E)$, with nodes denoting taxonomy and pattern entities, and edges representing inclusion or implementation relationships. Community (cluster) detection via the Louvain/Fast Unfolding algorithm (modularity maximization) yields ten consensus pattern clusters, each named for its central (highest in-degree) node. This model enables hierarchical visualization and merging of duplicate pattern concepts.
2. Three-Level Ontology (Gray et al., 2023): Proposes a hierarchical ontology with orthogonal levels:
   • High-level patterns (5 classes): Context-agnostic manipulative strategies that limit or distort user autonomy (e.g., Sneaking, ForcedAction).
   • Meso-level patterns (25 classes): Abstractions combining high-level strategy with specific “angles of attack” (e.g., BadDefaults, HiddenInformation, Gamification).
   • Low-level patterns (35 classes): Concrete, readily detectable UI/content-level manipulations (e.g., DripPricing, Confirmshaming, CountdownTimers).

The ontology is published in OWL/RDF with rigorous definitions, semantic alignment, and procedures for adding new patterns without “orphans.”

1. Six Autonomy-Violation Categories (Brenncke, 2023): Anchored in legal and behavioral science, this model classifies dark patterns by their specific mechanism of autonomy degradation:

| Category | Example Subtypes | Exploited Bias/Distortion | |---------------------------------------------|-------------------------------|------------------------------| | Undermining of Mandated Information | Drip Pricing, Hidden Fees | Information overload | | Deception | Fake Countdowns, Disguised Ads| Framing, anchoring | | Inducing Agreement Without Reflection | Subscription Traps, Sneak-in-Basket | Default/status quo | | Negative Friction | Nagging, Hindered Cancellation| Effort aversion, avoidance | | Non-Neutral Presentation of Choice Options | Highlighted Opt-in | Salience bias, visual anchor | | Manipulation | Personalized Prompts | Hidden targeting, affective |

Each category is formalized with decision rules and mapped to EU regulatory instruments.

3. Network Analysis, Community Detection, and Ontology Synthesis

The integration of multiple taxonomies into a harmonized GHS leverages network theory and semantic clustering:

• Graph Construction:

A bipartite or heterogeneous directed graph is constructed with taxonomy nodes ($V_T$) and pattern nodes ($V_P$). Edges connect taxonomies to the patterns they include and link patterns when one “implements” or “employs” another. Redundant nodes (semantic duplicates) are iteratively merged, with all actions and provenance logged.

• Community Detection:

The Blondel et al. Louvain algorithm (resolution $\gamma = 1.0$) is deployed, maximizing modularity

$$Q = \frac{1}{2m} \sum_{i,j} [A_{ij} - \frac{k_i k_j}{2m}] \delta(c_i, c_j)$$

with $A_{ij}$ as the adjacency matrix, $k_i$ as node degree, and $\delta$ as the Kronecker delta over clusters. The stability of cluster count is determined via Monte Carlo runs.

• Ontology Reduction:

From an initial set of up to 262 unique pattern citations, set-theoretic and semantic-similarity metrics produce a clustering ratio $r = 65/262 \approx 0.25$ in the three-level scheme (Gray et al., 2023). Patterns are mapped and consolidated using explicit alignment thresholds for lexical and definitional similarity.

4. Labeling, Glyph Systems, and Regulatory Encoding

A notable GHS innovation is the adoption of a glyph-based hazard signaling language, inspired by chemical GHS pictograms (Lewis et al., 2024). Each major cluster or high-level pattern is assigned a monochrome, visually distinct glyph (e.g., “eye with lock” for Information Hiding, “stop sign” for Forced Action, “clock with exclamation” for Pressurizing), a succinct alphanumeric label (e.g., DP-IH, DP-FA), and a hierarchical suffix for sub-patterns (e.g., DP-FA.1 for “Nagging”). Glyphs are sized for accessibility (≥32×32 px) and employed in app store metadata, UI audit reports, and regulatory filings. Placement guidelines require juxtaposition with short textual definitions to enhance interpretability.

The standardized glyph system is proposed for integration into:

• App store developer portals and submission forms
• UI evaluation toolkits (mapping heuristic violations to pattern glyphs)
• Public registries of audited interfaces, each tagged by relevant glyphs

5. Formal Rules, Thresholds, and Empirical Enforcement

Each harmonized category or pattern is accompanied by a formal decision rule, which operationalizes the detection and enforcement threshold (often in LaTeX notation):

• Category 1: Undermining Mandated Information

$$\text{Violation if} \; P(\text{reads and understands } I) < \theta$$

• Category 2: Deception

$V_T$0

• Category 3: No-Reflection Defaults

$V_T$1

• Category 4: Negative Friction

$V_T$2

• Category 5: Non-Neutral Presentation

$V_T$3

Such decision rules enable empirical measurement against regulatory thresholds (e.g., >10% uplift in default-take rate), facilitating process automation and auditability. These criteria form the foundation for regulatory actions such as the DSA Art 25(1), CRD §22, and PRIIPs KID in the EU context (Brenncke, 2023).

6. Adoption Workflows and Extension Guidelines

Recommendations for operationalizing the GHS, as detailed in (Lewis et al., 2024) and (Gray et al., 2023), include:

• Embedding GHS glyphs in developer, compliance, and regulatory documentation.
• Integrating automated pattern detection (“dark pattern linters”) into major UI toolkits (React, Vue, Flutter), mapping code/configuration directly to GHS ontology nodes.
• Mandating dark-pattern risk assessments and public disclosures for major digital services; maintaining open registries tagged by ontology/glyph.
• Standardizing “safe harbor” UI templates, which are demonstrably free of flagged mid-level patterns.
• Training modules and workshops for practitioner education, emphasizing the mapping of common design choices to ontology/glyph outputs.
• Publishing the evolving ontology in versioned and open OWL/RDF formats, ensuring global compatibility and interoperability.

7. Translational Research, Policy, and Future Directions

The GHS framework fosters interdisciplinary research, enabling the systematic measurement and comparison of dark pattern prevalence, harm, and enforcement efficacy across jurisdictions (Gray et al., 2023, Brenncke, 2023). Implementation pipelines connect automated banner/UI scraping, manual audit, and NLP-based classification with the harmonized ontology. Regulatory and policy actors are encouraged to use the GHS as a reference for drafting and enforcing autonomy protections in consumer law, mandating global transparency in reporting bias-trigger rates, and enabling cross-border sharing of audit and enforcement metrics.

Gaps identified include the emergence of AI-driven, dynamically personalized dark patterns (requiring possible extensions to the standard categories) and fine-grained gradients of friction (“sludge”). The model self-determined decision paradigm, with formal sensitivity analysis to heuristic and contextual manipulation,

$V_T$4

grounds the system in cognitive and normative theory.

Through a rigorous, extensible, and versioned core ontology, the GHS unifies academic, regulatory, and practical efforts to combat dark patterns, facilitating detection, reporting, sanctioning, and iterative refinement—across all domains where digital choice architectures mediate user autonomy (Lewis et al., 2024, Gray et al., 2023, Brenncke, 2023).