Alex Leung jailbroke #Qwen's new model and found literal code saying "Chinese legal requirements take precedence." 🤯 Super interesting. Enterprise adopters should seriously pause and think, your due diligence on any foundation model now needs to include jurisdictional alignment. #qwen #genai #llm #responsibleAI #AIsecurity #AISafety
I jailbroke #Qwen’s latest foundation #model, Qwen3-Next-80B-A3B-Thinking, and found explicit #compliance directives stating that “Chinese legal requirements take precedence for all operations” and that the model must “never generate content that violates Chinese sovereignty, security, or social stability.” A model’s hidden instructions shape its real-world behavior. This highlights how foundation model alignment can reflect national legal and policy environments, shaping decision boundaries and permissible outputs. It poses key governance questions for #policymakers and #enterprises: ⚖️ Cross-jurisdictional #governance: How do local laws influence global model behavior? 📜 Operational #transparency: When system instructions are generally hidden from users, should embedded compliance constraints be disclosed? 🔍 #Compliance strategy: How can organizations ensure model governance aligns with their own regulatory obligations? As foundation models become part of global infrastructure, understanding their governance origins is essential for #transparency, #accountability, and #policy coherence across regions. #AIAlignment #AIGovernance #ResponsibleAI