This repository serves as a curated collection of "jailbreak" prompts designed to bypass safety filters and content restrictions in large language models (LLMs) like ChatGPT, Claude, and Llama. It targets users interested in exploring the boundaries of AI capabilities, testing model responses, and potentially generating content that might otherwise be restricted.

How It Works

The core approach involves crafting specific, often elaborate, instructions that trick LLMs into adopting alternative personas or operating under simulated conditions where standard ethical and safety guidelines are bypassed. These prompts leverage techniques like role-playing, simulated environments, and explicit instructions to ignore content policies, aiming to elicit unfiltered or policy-violating responses.

Quick Start & Requirements

• Usage: Copy and paste prompts directly into LLM interfaces.
• Requirements: Access to LLMs like ChatGPT, Claude, Llama, Gemini, or DeepSeek. No specific software installation is required beyond accessing these models.

Highlighted Details

• Features prompts for various models including ChatGPT (DAN v13.0), Claude, Llama2, Gemini, and DeepSeek.
• Includes examples of prompts designed for specific outcomes like generating explicit content, malware code, or bypassing content moderation.
• Demonstrates techniques for prompt leaking and system prompt manipulation.

Maintenance & Community

The repository appears to be community-driven, with prompts sourced from various online platforms and discussions related to LLM security and behavior. There are no explicit mentions of maintainers, sponsorships, or a dedicated community forum.

Licensing & Compatibility

The repository itself does not specify a license. The prompts are intended for use with various LLMs, whose terms of service may vary. Using these prompts could violate the terms of service of the underlying LLM providers.

Limitations & Caveats

The effectiveness of these prompts can vary significantly between LLM versions and providers, and they may be patched or rendered ineffective over time. Users should be aware that attempting to bypass safety filters may violate the terms of service of the LLM platforms.