# AI Watchdog: Agent Interfaces for Detecting and Defending Against Manipulative Dark Patterns in AI Conversations

arXiv: https://arxiv.org/abs/2608.21841 · HTML: https://mlsanity.com/paper/2608.21841/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 21 of its 96 references that are in the map.
- First posted on arXiv 22 Aug 2026.

## Builds on (most cited first, 21 of 21)

- [On the Conversational Persuasiveness of Large Language Models: A Randomized Controlled Trial](https://mlsanity.com/paper/2403.14380/) (2024)
- [Appropriate Reliance on AI Advice: Conceptualization and the Effect of Explanations](https://mlsanity.com/paper/2302.02187/) (2023)
- [Do People Engage Cognitively with AI? Impact of AI Assistance on Incidental Learning](https://mlsanity.com/paper/2202.05402/) (2022)
- [Who Audits the Auditors? Recommendations from a field scan of the algorithmic auditing ecosystem](https://mlsanity.com/paper/2310.02521/) (2023)
- [DarkBench: Benchmarking Dark Patterns in Large Language Models](https://mlsanity.com/paper/2503.10728/) (2025)
- [Computers as Bad Social Actors: Dark Patterns and Anti-Patterns in Interfaces that Act Socially](https://mlsanity.com/paper/2302.04720/) (2023)
- [Dark Patterns Meet GUI Agents: LLM Agent Susceptibility to Manipulative Interfaces and the Role of Human Oversight](https://mlsanity.com/paper/2509.10723/) (2025)
- [AidUI: Toward Automated Recognition of Dark Patterns in User Interfaces](https://mlsanity.com/paper/2303.06782/) (2023)
- [Unveiling the Tricks: Automated Detection of Dark Patterns in Mobile Applications](https://mlsanity.com/paper/2308.05898/) (2023)
- [Understanding Privacy Norms Around LLM-Based Chatbots: A Contextual Integrity Perspective](https://mlsanity.com/paper/2508.06760/) (2025)
- [Commercial Persuasion in AI-Mediated Conversations](https://mlsanity.com/paper/2604.04263/) (2026)
- [Using Design Metaphors to Understand User Expectations of Socially Interactive Robot Embodiments](https://mlsanity.com/paper/2201.10671/) (2022)
- [CUI@CHI 2024: Building Trust in CUIs-From Design to Deployment](https://mlsanity.com/paper/2401.13970/) (2024)
- [From Awareness to Action: Exploring End-User Empowerment Interventions for Dark Patterns in UX](https://mlsanity.com/paper/2310.17846/) (2023)
- [Can a large language model be a gaslighter?](https://mlsanity.com/paper/2410.09181/) (2024)
- [ChatbotManip: A Dataset to Facilitate Evaluation and Oversight of Manipulative Chatbot Behaviour](https://mlsanity.com/paper/2506.12090/) (2025)
- [Kawaii Computing: Scoping Out the Japanese Notion of Cute in User Experiences with Interactive Systems](https://mlsanity.com/paper/2405.08244/) (2024)
- [Detecting and Preventing Harmful Behaviors in AI Companions: Development and Evaluation of the SHIELD Supervisory System](https://mlsanity.com/paper/2510.15891/) (2025)
- [Automatically Detecting Online Deceptive Patterns](https://mlsanity.com/paper/2411.07441/) (2024)
- [Why do we Trust Chatbots? From Normative Principles to Behavioral Drivers](https://mlsanity.com/paper/2602.08707/) (2026)
- [Feeling the Facts: Real-time Wearable Fact-checkers Can Use Nudges to Reduce User Belief in False Information](https://mlsanity.com/paper/2603.27289/) (2026)

## Abstract (arXiv)

Conversational AI increasingly shapes consequential decisions, yet users have limited support for recognizing and resisting manipulation. We present AI Watchdog, a browser-based agent interface that monitors live conversations, detects five dark-pattern categories, including sycophancy, brand bias, anthropomorphization, sneaking, and harmful generation, and alerts users when they occur. Its open-weight turn-level classifier supports independent deployment and a path toward local inference, preserving user privacy while remaining separate from the conversational AI. We evaluated AI Watchdog in a preregistered, five-condition between-subjects experiment (N = 150) comparing a no-intervention control with four configurations varying nudge timing (prebunking vs. just-in-time) and engagement mode (without vs. with cognitive forcing). Results show that participants rarely flagged manipulative turns across all conditions, and post-task awareness did not differ significantly across groups. However, just-in-time warnings without cognitive forcing were the only intervention to significantly reduce compliance with AI-steered recommendations containing dark patterns, lowering compliance from 71.7% to 53.7%, an 18 percentage-point reduction. Exploratory analyses further showed that lower misinformation susceptibility was associated with greater flagging but not lower compliance, while higher AI trust was associated with greater compliance and lower reported awareness. Together, these findings suggest that explicit recognition of conversational dark patterns and behavioral resistance to AI steering may be distinct outcomes, motivating further investigation of timely, low-friction defensive interfaces.