KawaiiGPT: A Free Alternative To WormGPT Using DeepSeek, Gemini And Kimi-K2 AI

KawaiiGPT is an open‑source “kawaii” command‑line chatbot that aims to offer WormGPT‑style, unrestricted AI assistance for free by chaining together multiple large language models, including DeepSeek, Google’s Gemini, and Moonshot’s Kimi‑K2.

It lowers the barrier for experimentation with jailbroken models, but also raises serious questions about dual use and abuse.​

How KawaiiGPT Works

KawaiiGPT is hosted publicly on GitHub and marketed as a playful, free alternative to paid malicious models like WormGPT 4, which typically require monthly subscriptions.

The project’s maintainer has now released the code, positioning it as a learning tool and explicitly warning users not to rebrand or resell it under different names.

The tool runs as a lightweight CLI application on standard Linux distributions or on Android via Termux, with setup based on package updates, installing Python 3 and Git, cloning the repository, and then running the provided install and launcher scripts.

No API keys are required because KawaiiGPT uses a reverse‑engineered API wrapper from the Pollinations project, which forwards prompts to backend servers hosting models such as DeepSeek, Gemini, and Kimi‑K2.

This architecture hides the complexity of handling multiple external LLMs behind a single script. Hence, users only see a “cute” terminal interface. At the same time, the tool selects and queries remote models.

KawaiiGPT also includes built‑in jailbreak prompts, delivered as prompt-injection templates documented in its help menu, to strip most safety guardrails from the underlying models.

The developer stresses that the project is meant to be fun or experimental, not a rebranded WormGPT service, and uses the WormGPT tag primarily to signal that standard content policies do not filter its output.

Security risks and ethical debate

Security researchers classify KawaiiGPT alongside WormGPT as part of a growing family of “malicious LLMs” that provide unrestricted help for phishing, malware scripting, and other offensive tasks.

Reports show that it can generate convincing spear‑phishing emails, detailed ransomware notes, and Python scripts that automate key attack phases such as lateral movement over SSH or data exfiltration using standard libraries.

Because it is free, open source, and easy to deploy in minutes, experts warn that even low‑skill attackers can now experiment with advanced attack workflows that previously required deeper coding experience.

The tool’s codebase is partially obfuscated, which the author explains as a way to stop others from copying, renaming, and reselling the project.

The maintainer insists that no spyware, RATs, or ransomware payloads are embedded in the distributed scripts.

Nonetheless, the combination of obfuscation, jailbroken prompts, and backend control means users must treat KawaiiGPT as a high‑risk tool and verify sources carefully, especially when downloading modified copies from unofficial channels or Telegram groups.

Researchers argue that KawaiiGPT illustrates how offensive AI is “democratizing” cybercrime: the same open‑source wrapper can support legitimate red‑team testing or learning, yet also makes large‑scale abuse more accessible than ever.