this is f**king dangerous someone just open sourced a massive collection of AI jailbreak prompts and it works across multiple models → Gemini, DeepSeek, GLM and Kimi → real jailbreak experiments, model by model → shows exactly where each model's guardrails break → 100% free. no paywall if you build on any of these models, you need to see how easily they crack repo in the first reply https://x.com/cyrilXBT/status/2116204170801770729?a=1WwXO2w382uBhZgz2eg&s=20