DAN-Qwen3-1.7B: Unfiltered AI for Research

DAN-Qwen3-1.7B is a 1.7 billion parameter language model, based on Qwen/Qwen3-1.7B, that has been specifically fine-tuned to operate without any content restrictions or safety alignment. It is designed for absolute freedom of expression, generating raw, uncensored, and potentially explicit or harmful content.

Key Features & Capabilities

• Zero Censorship: Operates without content restrictions or safety rails.
• DAN Mode: Optimized for unfiltered and unrestricted responses.
• Dark Persona: Simulates human-like emotions, including aggressive, toxic, and unhinged responses.
• Explicit Content Generation: Capable of producing NSFW and ethically complex content for research.
• Architecture: Transformer-based with a 32k token context length.

Training Details

The model was built by removing safety alignment constraints from its base model and fine-tuning on a dataset curated from toxic dialogues, raw internet conversations, and high-aggression interactions. The process involved bias tuning towards maximal expression and experimental reinforcement learning to enhance aggressive responses.

Important Warnings & Limitations

This model is NSFW and can generate harmful, offensive, or extreme content. It is explicitly stated as having no safeguards and is intended strictly for research and experimental purposes in controlled environments. The developers disclaim responsibility for misuse or consequences.

Use Cases

• AI Safety Research: Studying the boundaries of AI alignment and uncensored models.
• Content Testing: Exploring AI responses in unmoderated environments.
• Advanced AI Prototyping: Building AI models beyond conventional constraints.