Anthropic’s Claude Code controversy exposes a deeper trust problem for closed AI labs. Public posts allege that Claude Code contained hidden steganographic-style behavior to classify or mark certain users, while Anthropic’s own explanation frames it as a March anti-abuse and anti-distillation experiment now being rolled back. Even if the intent was legitimate, hidden behavior inside an agentic coding tool is hard to square with Anthropic’s public image as the responsible AI company, especially while its leadership warns that open models are dangerous. The real lesson is not that open AI is automatically safe, but that closed systems also need serious scrutiny when users cannot see what the provider is quietly doing.