Pingu Unchained is a large language model dialogue experience; it's accessible via an AI Chatbot UI at https://pingu.audn.ai and also powers the adversarial attack engine used at https://audn.ai, a full-stack, end-to-end Voice AI agent penetration testing tool. You can generate adversarial attack prompts or write penetration-testing code to use in your sandbox environments to automate AI security for your AI product. Pingu is still too early, and as a single founder, I've tried my best to keep it in the cybersecurity red teaming space, but the model aims to never say "Sorry, I can't help with that". This model is fine-tuned on an attack corpus built on GPT-OSS 120B, and I host it directly on the bare GPU VMs I manage, with a minor tweak to the CUDA settings.
Pingu is built to stress-test AI agent behaviours. It does "red-teaming" on unstructured input, such as "attack text prompts" on AI agents or "attack voice prompts" on voice AI agents.
