Anthropic's Smart Reports let managers audit team Claude use, and the real question is who controls the data

Anthropic has added a feature called "Smart Reports" to its enterprise offering. Futurism reports that, according to Anthropic's own support page, the tool lets managers generate summaries of "how a team uses Claude" in order to "report on the work getting done." The page also says managers can sort a team's AI activity by overall cost and by what the chatbot produced. That gives them a direct view of how efficiently each team appears to use the tools.

Anthropic says sessions are anonymized. Futurism is skeptical. It points to what it calls a previous Anthropic "privacy disaster," in which, according to the outlet, supposedly anonymized data was easily re-identified using common large language models. Futurism's headline goes further and calls the feature a way for "your boss to spy on your Claude chats." That is the outlet's framing. Based on the support page as described, the stated purpose is team-level reporting, not reading individual conversations. How well the anonymization holds up in practice has not yet been shown, and that is the part to watch.

The business context explains the timing. Futurism reports that corporate customers make up about 80% of Anthropic's business, with consumer subscriptions making up the rest. With an IPO on the horizon, Anthropic's valuation depends mostly on winning and keeping enterprise contracts. Companies that pay for Claude across a whole organization want to know what they are getting for the money. Smart Reports answers the question every CFO is asking: is this spending producing work? Viewed that way, it is a return-on-investment dashboard as much as a monitoring tool.

This fits a pattern we have been following. Inside companies, AI is moving from a separate product to the normal way work gets done. Anthropic recently merged chat and its Cowork agent mode in Enterprise, so the assistant runs code and handles files as part of ordinary conversation. Once most of someone's working day runs through an AI system, the logs of that use become a detailed record of the work itself. In theory, a manager could learn more about an employee from those logs than from any timesheet or email archive. The data was always going to exist. What is new is that the vendor now turns it into a manager-facing product.

Our reading: we would not call this sinister, and we would not call it harmless. In the short term the risk is real. Sorting employees by "cost" and "output" encourages crude productivity scoring. It can penalize the careful worker who uses AI less but uses it well. It can also discourage people from experimenting, which is exactly what companies need from them right now. Anonymization only protects workers if it is strong and independently checked, and Futurism's reminder about re-identification is a fair warning. In the long term, measuring how AI changes work is necessary. Without it, companies cannot tell which tasks to hand to software and which to keep for human judgment, and that distinction drives the shift we expect toward more meaningful work. What decides the outcome is governance. Can employees see what is reported about them? Is the reporting kept at team level by design? Is there an audit trail showing who looked at what? Do works councils or unions have a say? As with the data-sovereignty offerings we have covered, the question to ask is who controls the perimeter of the data, not what the dashboard looks like.

What to watch next is whether Anthropic publishes technical detail on how the anonymization works, whether regulators (especially in Europe, where workplace monitoring is tightly restricted) look into tools like this, and whether competitors ship similar features. Our bet is that they will. Tools that measure workplace AI use are about to become standard. The useful fight now is over the rules for that measurement, because blocking the tools outright is unlikely to work.