DarkLens - Dark Pattern Evidence Auditor

DarkLens scans the page you're currently viewing for evidence of potentially manipulative interface patterns, commonly known as "dark patterns," and documents what it finds. It does not determine legal liability and is not a compliance tool. WHAT IT DETECTS (V0.2) DarkLens uses rule-based detection (pattern matching + DOM inspection, with no AI/LLM classifier) to identify three pattern classes: • Confirmshaming: Guilt-based or manipulative language associated with decline or opt-out actions. • Preselection: Potentially non-essential options that are selected by default. • Visual Interference: Interface designs that may obscure or de-emphasize a meaningful choice. Detection is scoped to relevant UI containers to reduce false matches across unrelated page sections. HOW IT WORKS Every finding is assigned an evidence tier rather than a vague confidence score, making the strength and limitations of the detected signal explicit. When DarkLens lacks sufficient evidence to evaluate a pattern reliably, it can flag the case for manual review rather than silently treating it as safe or making an unsupported conclusion. Findings may be mapped to a static, dated regulatory reference table for research purposes. These mappings indicate potentially relevant provisions and are never presented as legal conclusions. PRIVACY • Detection runs locally in your browser. • No page content is sent to a server. • No tracking or analytics. • No account required. • No LLM or external AI model is used in V0.2. WHY THE PERMISSIONS • activeTab / scripting: Required to inspect the DOM of the webpage being analyzed and run local detectors. • storage: Required to retain findings for display in the side panel. • Host access: Allows DarkLens to analyze webpages across different sites. DarkLens does not transmit page content to external servers. WHO THIS IS FOR DarkLens V0.2 is an open-source research and portfolio prototype exploring dark-pattern detection, digital consumer protection, AI/UX governance, responsible technology, AI ethics, and adversarial testing methodologies. It is not a finished commercial compliance product. End-to-end benchmark evaluation is still in progress, so V0.2 makes no established claims about precision, recall, or overall detection accuracy. Known limitations and failure modes are documented in the project's SYSTEM-CARD. Open source. Rule-based. Privacy-first. Evidence, not verdicts.