Anthropic constantly uses dark patterns to steal training data from customers (like the “how is claude doing” spam, data retention loosening when the safeguards false positive, etc).

How is that a dark pattern? What is the light pattern for getting feedback from users?