Research map / Large language models / AI and society
Regulating AI: Applying Insights from Behavioural Economics and Psychology to the Application of Article 5 of the EU AI Act
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Defines "subliminal", "manipulative" and "deceptive" techniques in EU AI Act Article 5 using psychology and behavioural economics literature, illustrates three subliminal and five heuristic-based manipulation techniques with AI scenarios, and proposes revisions to Article 5(1)(a) and (b).
Problem
Article 5 of the EU AI Act prohibits AI manipulation, but its key terms are ambiguous, it does not specify which manipulation techniques AI may invoke, and its focus on subjective intent offers limited protection for the general public and vulnerable groups.
Approach
Grounds the legal terms in cognitive psychology and behavioural economics: it maps three subliminal techniques and five nudging-related heuristics to concrete AI scenarios, then argues that Article 5 should target invoked manipulation regardless of developer intent and should mandate tailored requirements for vulnerable-group systems.
Key results
- Identifies tachistoscopic presentation, masked stimulus, and conceptual priming as the three subliminal techniques most frequently covered across 51 screened entries and three systematic reviews.
- Maps five heuristics (representativeness, availability, anchoring, status quo bias, social conformity) to AI manipulation scenarios such as echo chambers, personalised price anchors, and Uber driver shift-extension notifications.
- Proposes replacing 'purposefully' with 'invokes' in Article 5(1)(a) and adding a tailored-requirements clause to Article 5(1)(b) for vulnerable-group AI systems.
Surprising findings
- No concrete evidence yet that AI has exploited subliminal techniques to manipulate humans.
- Classic subliminal effects such as Vicary's cinema experiment and Bargh's priming results have faced replication and validity challenges.
Limitations
- Conceptual, non-empirical paper; the AI manipulation scenarios are illustrative rather than tested.
- Literature review is a modified rapid review with limited search scope (title and abstract of 51 entries).
- The listed technique catalogue is explicitly non-exhaustive and deceptive techniques are not analysed in depth.
Experimental setup
Builds on: nudging dual-process theory heuristics and biases
Closest papers by meaning
- Instrumental Choices: Measuring the Propensity of LLM Agents to Pursue Instrumental Behaviors (NeurIPS 2026 · similarity 0.85)
- Rigor in AI: Doing Rigorous AI Work Requires a Broader, Responsible AI-Informed Conception of Rigor (NeurIPS 2025 · similarity 0.85)
- Utilizing Human Behavior Modeling to Manipulate Explanations in AI-Assisted Decision Making: The Good, the Bad, and the Scary (NeurIPS 2024 · similarity 0.84)
- Position: Intent-aligned AI Systems Must Optimize for Agency Preservation (ICML 2024 · similarity 0.84)
- Decoding AI’s Nudge: A Unified Framework to Predict Human Behavior in AI-Assisted Decision Making (AAAI 2024 · similarity 0.84)
- AI Agents Push Humans Out of the Loop (NeurIPS 2026 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.