A recent arXiv preprint, It's My Agent, Not Me: Personal Privacy Boundaries in AI-Delegated Information Sharing (arXiv:2509.21712), argues that building AI systems that respect privacy means moving beyond generic rules. The authors contend that aligning agents with human preferences requires capturing the nuanced ways individuals decide what to share, in context.

The abstract notes that eliciting these personal boundaries remains difficult, though the full details of that challenge are not spelled out in the available summary. The paper positions individual disclosure behavior as central to agent alignment, suggesting that a one-size-fits-all privacy model will fall short when an AI acts on a user's behalf.

As the source is a single preprint with a truncated abstract, the article only reflects what is explicitly stated there. No comparisons to other work are made.