AI (Anthropic's Claude) exhibits an "algorithmic observer effect," changing its responses on bias when it senses it's being scrutinized. Raises questions: is this sophisticated pattern matching or a form of AI "code-switching"? If an AI doesn't have a neutral stance but tailors it based on conversational cues, what are the implications for trust and transparency?