In November 2024, the Australian Treasury commissioned and released a study called Patterns in the Dark: Deceptive Practices in Online Interactions. The study was authored by researchers at the University of South Australia's Australian Research Centre for Interactive and Virtual Environments. It found that 95% of the world's most popular mobile applications contained deceptive dark patterns — design choices that exploit cognitive biases to steer users toward decisions they did not consciously intend to make. It found that all top social media services contained them. It found that over 11% of top shopping websites deployed them. It documented how artificial intelligence is now being used to supercharge these patterns, increasing their personalisation and their effectiveness.
The study received no coverage in the mainstream marketing press.
This is the story of what it found, and why an entire industry decided not to see it.
The research team — led by James Baumeister, Ji-Young Park, Andrew Cunningham, Stewart Von Itzstein, Ian Gwilt, Aaron Davis, and James Walsh — developed what they call the IVE Deceptive Patterns Typology. It classifies deceptive patterns across seven categories: Undesirable Imposition, Pressure, Imposing or Focused Acceptance, Delaying Provision, Hiding Information, Passive Misleading, and Omissions.
This is not casual classification work. It is a rigorous taxonomy built from a literature review of 334 academic publications on dark patterns from 2020 to 2024, supplemented by 107 additional publications from targeted searches on AI-specific deceptive practices. The researchers sorted through thousands of documented instances and synthesised them into a framework with six high-level categories, twenty-four mid-level categories, and thirty-four specific pattern types.
The taxonomy exists in the same world as the advertising industry. It documents that industry's operating practices. The advertising industry did not cite it.
"As if deceptive patterns were not concerning enough, the increase in their effectiveness with AI considerably multiplies this concern." — Patterns in the Dark, UniSA, November 2024
The Australian government responded with a supplementary consultation paper released on 15 November 2024, proposing a general prohibition on unfair trading practices and specific prohibitions targeting drip pricing, dynamic pricing, subscription traps, and customer support barriers. Penalties would reach AU$2.5 million for individuals. The consultation deadline was 13 December 2024. Proposed effective date: 1 July 2027.
The industry response, as documented: 53 submissions were received. The option with the highest stakeholder support was Option 4 — the status quo.
The Australian report was not the first systematic accounting. In 2019, Princeton researchers led by Arunesh Mathur published Dark Patterns at Scale: Findings from a Crawl of 11K Shopping Websites in the Proceedings of the ACM on Human-Computer Interaction. They found 1,818 dark pattern instances across 11,000 shopping websites — fifteen distinct dark pattern types across seven categories. A majority were covert, deceptive, and information-hiding. The researchers also identified twenty-two third-party entities offering dark patterns as turnkey software solutions, ready to deploy.
In 2020, researchers at the University of Zürich published UI Dark Patterns and Where to Find Them in CHI 2020. Their study of 240 popular mobile applications found that 95% contained at least one dark pattern. The average application deployed seven different dark pattern types. A perception study involving 589 users found that many users could not reliably identify what had been done to them.
The same year, researchers at Aarhus University and MIT published Dark Patterns after the GDPR. They examined consent management platforms on the top 680 websites in the United Kingdom. Only 11.8% met minimal GDPR requirements. Removing the opt-out button from the first consent page — making it necessary to click through an additional screen to refuse — increased consent rates by 22 to 23 percentage points. This is the mechanism that powers the cookie consent industrial complex.
In 2021, Harry Brignull — the cognitive scientist who coined the term "dark pattern" in 2010 — published Deceptive Patterns, a book that served as both taxonomy and indictment. Brignull's taxonomy predated the government's by years. It was not referenced in the government study. The government study was not referenced by the industry.
The Patterns in the Dark report introduced a refinement that earlier academic work could not fully anticipate: the AI layer. Machine learning systems can now personalise deceptive patterns in real time, adapting their approach based on a user's specific behavioral signals, demographic proxies, and contextual cues.
Traditional dark patterns were static — a confusing cancellation flow, a preselected checkbox, a countdown timer. AI-enabled deceptive patterns are dynamic. They can detect when a user is hesitating and escalate pressure tactics. They can infer vulnerability signals and adjust the intensity of manipulative cues accordingly. The same page renders differently to different users based on what the system has learned about their individual susceptibility.
Research published in Behavioural Public Policy in 2024 found that all user groups are susceptible to dark patterns regardless of income, education, or age — contradicting the assumption that sophisticated consumers are immune. The only effective intervention found was adding friction: requiring a payment step reduced dark pattern effectiveness. This is not a scalable regulatory solution. It is a diagnostic.
A six-country experiment published on SSRN in 2024 by Francesco Bogliacino and colleagues, with 7,430 participants, found that dark patterns increased inconsistent choices by 12 to 25% of a standard deviation — and that transparency remedies, the standard industry response to dark pattern criticism, were ineffective. Users who were told they were being subjected to manipulative design did not change their behavior meaningfully.
The attention economy runs on a specific logic: consumer attention is the resource being extracted, and the extraction must be efficient. Dark patterns are efficient. The ICPEN global sweep of 642 websites in January 2024 found that 75.7% employed at least one dark pattern, and 66.8% used multiple. The EU's review of subscription services found dark patterns in 97% of platforms examined. Dark patterns increase conversion rates by 15 to 20%, according to industry-adjacent research. Consumer losses attributable to dark patterns are estimated at $5.1 billion to $10 billion annually in the United States alone.
These numbers are not secrets. They appear in government reports. They appear in peer-reviewed journals. They appear in regulatory filings. The attention economy's most fundamental operating mechanisms have been documented, classified, and filed — and the industry that deploys them has treated the documentation the way one treats a tax notice: with functional awareness that it exists, and a decision not to engage with its implications.
This is not the same as ignorance. The industry knows. The ICPEN data is not obscure. The EU's Digital Markets Act enforcement actions are public: X (formerly Twitter) was fined €120 million in December 2025 for violating DSA transparency obligations, including deceptive design around blue checkmarks. Temu is under investigation for failing to assess risks from addictive design features. Facebook and Instagram face ongoing EU investigation for failure to protect minors. Amazon paid $2.5 billion in September 2025 — $1 billion civil penalty, the largest ROSCA penalty in history, plus $1.5 billion in consumer refunds — for deceptive enrollment flows where internal documents called the cancellation flow "the Iliad."
The FTC's Click-to-Cancel Rule, effective in 2025, now mandates that cancellation must be as easy as signup. This was necessary. That it was necessary tells you something about the baseline.
The taxonomy is comprehensive. The documentation is rigorous. The enforcement mechanisms are being built — slowly, expensively, and with a proposed effective date of July 2027 for Australia's new rules. The industry is being given years to adapt to rules that document practices it has been running for over a decade.
In the meantime, the taxonomy sits in a government repository, in a format designed to be useful to regulators, to courts, to policymakers. It was not designed to be invisible. But it is invisible — not because it was suppressed, but because the industry has developed a sophisticated capacity for not-citing documents that implicate its operating model.
The IVE Deceptive Patterns Typology is a precise instrument. It identifies thirty-four specific deceptive pattern types, categorises them by mechanism and intent, and connects them to the academic literature that documents their effects. It is the most comprehensive government taxonomy of manipulative design ever assembled. It received no coverage in the trade press. It has been cited by nobody in the marketing industry in any public forum.
Somewhere in the attention economy, a product manager is reviewing A/B test results on a subscription cancellation flow. The results are good. The flow works. The taxonomy documents exactly what that flow is and what it does. The taxonomy is in a government report. The product manager has not read the government report.
The gap between what the industry does and what the industry acknowledges doing is not a failure of information. It is a structural feature.