Anthropic built a tool so your boss can read your Claude chats. It calls this Smart Reports.

Anthropic has launched a feature called Smart Reports, which lets managers generate summaries of how their employees are using Claude at work. What they're doing, how much it costs, how productive the AI thinks they're being. The whole picture, served up for whoever holds the budget.

The feature is anonymous in theory. In practice, Anthropic's last attempt at anonymisation went badly: a February study found those sessions could be de-anonymised using off-the-shelf language models. So the comfort blanket has a known hole in it.

None of this is accidental. Corporate customers make up roughly 80 percent of Anthropic's revenue, and the company is heading towards an IPO. Smart Reports is not a product for the people typing into Claude. It is a product for the people paying for Claude, which is a distinction Anthropic seems increasingly comfortable making out loud.

The genuinely interesting question isn't whether this is surveillance. It obviously is. It's whether workers will now start treating Claude the way they treat work email: carefully, strategically, saying less than they mean.