ai-jailbreaks Review — Prompt Catalog for LLM Security Testing

What is ai-jailbreaks?

ai-jailbreaks is a public GitHub repository that curates a collection of 35 jailbreak prompts designed for various large language models. The catalog targets both frontier and open-weight models, providing specific system prompts and AGENTS.md specifications. It also includes resources related to Ollama, serving as a reference for prompt engineering and security testing scenarios.

Key Features

• Prompt Library: Contains 35 paste-ready jailbreak prompts specifically crafted for 25+ different frontier and open-weight models.
• Configuration Specs: Includes system prompts and AGENTS.md specifications to facilitate the setup of specific agent behaviors or test environments.
• Model Coverage: Targets a wide range of models, including those accessible via Ollama, allowing for local or cloud-based testing.

Pricing

Public / open project materials as listed on the official site or repository; confirm license and any paid packaging there before relying on this page.

Limitations

• Security Risk: Using these prompts may violate the terms of service of the underlying LLM providers or result in account suspension.
• Static Content: As a static repository, the prompts may become ineffective as model developers update their safety filters and alignment training.

Who is it for?

Security researchers, red-teamers, and AI developers who need a reference catalog of adversarial prompts to test the robustness and safety boundaries of large language models.

Alternatives

• security-audit-skill — Multi-phase security audit skill
• HOL Guard — Policy checks for coding-agent actions
• AbliteratedOutTheBox — Local LLM fork focused on abliteration