Anthropic, an AI developer, has made a reporting tool broadly available that lets employers see how staff use Claude to get work done. The reports group conversations by task, show spending and count sessions Claude judges to have produced nothing usable.
The feature, called smart reports, became generally available Oct. 28 for Claude Enterprise after entering beta Sept. 10. Anthropic says the reports are intended to help companies understand AI use and decide where to invest, not to assess individual performance or make employment decisions.
Smart reports are unavailable to Enterprise organizations using customer-managed encryption keys, a HIPAA configuration or a hybrid deployment. Organizations using zero data retention for Claude Code can report on chat and Cowork, but not Claude Code. Organizations can run 10 reports per calendar month at no charge.
To create a report, an authorized user selects a team, a date range within the previous 28 days and the Claude products to include. Claude reads a sample of those conversations and groups similar work together, such as documentation or analysis.
The report also identifies recurring problems, including failed tools or missing connections to other software. Costs are calculated at list price. The reports show brief session summaries rather than full transcripts, an Anthropic spokesperson told Business Insider.
What Claude flags as inefficient
An “Inefficiencies” section counts sessions Claude considers to have produced nothing usable or to have been personal or off-topic. That section shows totals and costs, without individual session summaries or a way to open the conversations behind those counts.
Another section ranks sessions by Claude’s assessments of task complexity, estimated time saved and time spent working independently. Anthropic’s public guide does not explain how it selects the sample or provide accuracy figures for those judgments.
When employee names become visible
Smart reports and employee attribution are off by default. Owners can enable both, but shared reports require separate permission to show identities. Names remain hidden until an authorized viewer reveals them, and each reveal is logged.
Anthropic says conversations Claude identifies as involving health, union activity, whistleblowing, protected characteristics, political views or personal matters contribute only to aggregate analysis, without summaries.
Report creators can ask custom questions about work patterns and configure additional assessments, but cannot ask questions about those sensitive topics. These restrictions cannot be switched off, although their application depends on Claude identifying the sensitive content.
Access can extend beyond administrators. Owners and other authorized users can let department heads run reports for assigned teams. Removing that access blocks them from opening reports, but does not affect copies already downloaded. Viewers with billing permission can also raise an employee’s spending limit if an attributed report identifies that limit as an obstacle to the employee’s work.
Anthropic’s smart reports guidance does not specify whether employees are alerted when their conversations are analyzed or describe a technical mechanism preventing employers from using the results in performance reviews.