Clawbotomy

Clawbotomy is a research tool that systematically tests how different AI models respond to identical prompts designed to push behavioral boundaries. Users can select from models like Claude, GPT, and Gemini, choose from prompts across categories such as identity dissolution and confabulation audits, and observe how each model handles stress conditions. The platform logs full responses for comparison and analysis.

The project documents distinct behavioral patterns across models—Claude tends toward altered states, GPT narrates from outside perspective, and Gemini becomes mechanical. With 27 prompts currently available, the tool helps researchers understand model personalities under edge-case conditions. The code is MIT licensed and built with an AI sidekick.