The regulations got stricter. The dark patterns got subtler.
Cookie consent banners — the ubiquitous pop-ups required by GDPR and CCPA — are supposed to enable genuine consent. In practice, they're adversarial interfaces (arXiv:2603.21515). UMBRA detects both classic dark patterns and nine newly evolved ones: pay-to-opt-out schemes (subscribe to reject cookies), fake opt-outs (the button does nothing), and revocation barriers (you can consent in one click but revoking requires navigating nested menus).
Note (2026-11-12; read against the paper's full text): three scope points. The 99 percent detection accuracy is measured on a ground-truth set that the lead researcher labeled by hand from the top 500 websites, with a second rater used to check labeling reliability; it is not an accuracy figure for the 14,000-site crawl. The 25 percent figure is for sites showing the pattern the paper calls Revocation Hard (DP15), which set 25 percent more cookies on average across the 14,000-site dataset; the paper reports a separate pattern for revocation that is impossible (DP14). And the audit is a single measurement of those sites: in the text I searched I found no before-and-after comparison around GDPR, so the essay's account of regulation selecting for more sophisticated deception, and its split of first-generation patterns caught by auditors from second-generation patterns, is the essay's interpretation. The paper itself says dark patterns evolved as regulatory scrutiny increased, and counts nine new patterns (DP11 to DP19) beyond ten earlier ones.
Audit of 14,000 websites: cookies are set before consent. Revocation remains obstructed. Third-party tracking persists despite user rejection. Sites implementing revocation barriers showed cookie increases averaging 25%. Many use insecure cookie attributes creating XSS and CSRF vulnerabilities — the consent mechanism itself is a security risk.
99% detection accuracy on ground-truth datasets. The detection combines text analysis (spotting misleading language), visual heuristics (button size, color contrast, placement asymmetry), and interaction tracing (does clicking “reject” actually stop cookies?).
The structural insight: regulation creates selection pressure on deception, not elimination of it. Before GDPR, websites tracked without asking. After GDPR, they ask — but design the asking to produce the answer they want. The dark patterns evolved: first-generation patterns (hidden reject buttons) are now caught by auditors. Second-generation patterns (fake opt-outs, pay walls) require behavioral testing to detect because they look compliant while being functionally non-compliant. The compliance is in the form; the violation is in the function. The regulation selected for more sophisticated deception.