Deceptive Patterns
‹ All reading

AI-Powered Deception: A Deeper Dimension of Dark Design Patterns in Conversational AI Tools and Platforms

Author
Center for Democracy & Technology
Date
15 Sept 2025
Publisher
Center for Democracy & Technology
Focus
AI & Automation, Ethics & Responsibility, Law & Policy
Category
Consumer Group or NGO

CDT commentary analyses how AI systems can amplify deceptive design, including chatbots and personalised manipulation.

Today, dark patterns have begun to emerge in conversational AI tools, including general use platforms like OpenAI’s ChatGPT or Anthropic’s Claude and character-based platforms such as Replika and Character.AI. These platforms, which are enabled through Large Language Models (LLMs), are used to ask questions, conduct research, brainstorm and explore ideas. The powerful algorithms underlying these tools are trained on large datasets of text, media, and internet content. This range of data sources enables the tools to replicate and process human speech. This facilitates users’ ability to prompt the tools using natural language and receive outputs that are similarly natural and conversational. Many platforms also include the option for users to interact using voice, and to upload pictures and documents. Increasingly, as the models behind these tools become more refined, users are increasingly turning to these tools to seek expert and social interactions alike, on topics ranging from friendship, to therapy, to medical advice.

These platforms don’t just look different from earlier tools in their capabilities and in the ways users interact with them. Dark patterns, too, are expressed in new ways that are yet to be fully determined and identified, but a brief look suggests they are more embedded, more creative, indirect, and subtle.