Anthropic is giving businesses a much closer look at how their staff use Claude, and the feature raises an obvious question: where does adoption analytics end and workplace surveillance begin?
The new tool, called smart reports, launched this week for organizations on paid Claude Enterprise plans and has to be switched on in an organization’s settings. A manager picks a team, a time window covering up to the previous 28 days, and which Claude products to include. Those can be standard chat, Claude Code, Claude Cowork, or the combined chat and Cowork experience. Claude then reads through the team’s conversation transcripts and sorts the activity into workstreams and output categories, such as vibe-coded projects or technical documentation, alongside a tally of what it all cost.
The output is a set of interactive charts. Clicking into a workstream shows individual sessions, which can be filtered and ranked by cost, product, date or output type. Managers can also ask their own questions about usage and spending and save them for later reports. The system offers suggestions too, such as new rules to add to a code repository. Admins with billing rights can raise a spend limit directly from the report if it appears to be slowing someone down. Each organization gets ten reports a month at no charge, and further reports cost extra.
The most sensitive part is a category Anthropic calls “inefficiencies.” It flags sessions that produced no usable output and sessions the system classifies as personal or off-topic. For employees, that means a chat assumed to be private could end up as a data point in a manager’s dashboard.
Anthropic says smart reports are meant for understanding adoption and planning investment, and that they should not be used to judge individual performance or make employment decisions. The tool also refuses queries about race, gender identity, religion, political affiliation, immigration status, health, union activity, harassment complaints, and employees’ personal lives.
Those safeguards matter, but they rely on how customers choose to behave. Anthropic can block certain questions, but it cannot control what a manager does after seeing that a session was labeled off-topic. That worry has real precedent. One recent survey found a quarter of US managers use AI often or always when deciding on layoffs. In July, former Meta employees claimed the company used poorly tested AI tools to track productivity and token consumption when choosing who to let go, an allegation Meta denies.
The timing is also notable. Days earlier, Anthropic’s IPO filing reported that existing customers had increased their spending by 900 percent year over year, along with an $80 billion loss for the first half of 2026. Tools that help enterprises justify and expand their AI budgets serve Anthropic’s commercial interests as much as its customers’. Whether workers will see these reports as a budgeting aid or as a monitoring tool is still an open question.
