Anthropic's Smart Reports Let Claude Enterprise Admins Analyse How Staff Use Claude

The feature, off by default, has Claude read a sample of employee transcripts and report on workstreams and spending, and it can reveal individual names if the organisation turns attribution on.

By Nakoda News Desk

2 days ago·3 min read

Prefer Nakoda News on Google

Anthropic has added a feature called Smart Reports to Claude Enterprise. It lets company administrators have Claude review a sample of employees' conversations and produce reports on how teams use the assistant and what that use costs. Critics say it gives employers a new view into staff chats, and Futurism ran its coverage on Tuesday under the headline "Anthropic Debuts Tool for Your Boss to Spy on Your Claude Chats."

Anthropic's help centre says Claude reads a sample of transcripts within a chosen scope and groups them into workstreams and output types. It then assigns spending at list price to each group and writes up findings, including friction points and repeated patterns that could be turned into shared skills. "Smart reports help you understand adoption and plan your investment in Claude," the documentation says.

Anthropic adds a warning in the same documentation: the reports "aren't designed and should not be used for evaluating individual performance or making employment decisions."

How It Works

Smart Reports is off by default, and only a Primary Owner or Owner can switch it on, according to Anthropic. Once it is on, Primary Owners, Owners, Admins and custom roles with analytics access can create and view reports. Access can also be delegated to team leads or department heads who are not administrators. Each report covers up to the last 28 days. Administrators can pick specific teams and Claude products and add custom questions, Gadget Review reported. Organisations get up to 10 reports a month at no charge, and more require contacting Anthropic's account team.

The documentation says reports show summaries of individual sessions, and that is where the privacy argument starts. A separate setting, "Allow attribution to individual users," decides whether names can be seen. With the setting off, drilldowns show "User" in place of names and hide session IDs. With it on, names stay hidden by default, but viewers can reveal names and session IDs from inside the report, and Anthropic says "Unveils are logged."

Some organisations cannot use the feature. Those with customer-managed encryption keys, a HIPAA configuration, or a hybrid deployment that runs inference on their own cloud are excluded. Where Claude Code is set to zero data retention, reports cover chat and Cowork but leave out Claude Code.

Topics It Will Not Drill Into

Anthropic says two safeguards are always on and cannot be switched off. Custom questions cannot target restricted topics. Sessions flagged as restricted feed only aggregate analysis, with no summaries. The restricted list includes protected characteristics, health and medical matters, union activity, harassment or whistleblower matters, political or religious views, personal life, and sentiment about named people.

Gadget Review noted that the flagging is automated and that Anthropic has published no data on how often it fails. Some sensitive conversations could therefore be put in the wrong category. The publication also said companies may need legal review of consent, notice and data-protection obligations before turning the feature on, because these differ between jurisdictions.

Why It Matters

As AI spending grows, chief financial officers and technology leaders want to know which teams get value from enterprise AI licences. Smart Reports gives them a way to find out, using conversation content rather than simple usage counts. For HR and legal teams, especially in jurisdictions with strict employee-monitoring rules, the attribution and drilldown settings are what will need a decision. Workers using Claude at work should assume their employer may be able to see summaries of their sessions.