Ethics of AI-generated content in production applications

As more companies ship AI-generated content to end users, I want to discuss the ethical considerations:

1. Disclosure: Should users always know they're reading AI-generated text? 2. Accuracy: What's the acceptable error rate for different domains? 3. Attribution: If an AI summarizes multiple sources, how should sources be cited? 4. Bias: How do you audit for bias in production AI outputs?

I'm writing a paper on this topic and would love to hear from practitioners dealing with these questions daily.