Your company pays for Claude, and starting now it can read the transcripts of what your team does with it. Anthropic launched Claude smart reports for Enterprise customers, and they show managers exactly how AI time gets spent. One detail is making people nervous: sessions get flagged as “personal or off-topic.”
What actually launched
Smart reports rolled out on October 28 for Claude Enterprise customers. An admin picks a team, a date range up to 28 days, and which products to include: chat, Claude Code, Claude Cowork, or the combined chat-and-Cowork experience. Claude then reads a sample of the team’s conversation transcripts and turns them into charts. The full breakdown is in Anthropic’s announcement and its support guide.
Scope has limits, though. Claude Duet, the proactive agent Anthropic announced in October, doesn’t factor into any of this because it isn’t available on Enterprise plans at all.
Three things come out the other end. First, workstreams: your week of Claude sessions gets grouped into buckets like “vibe coding” or “technical documents.” Second, deliverables, meaning the outputs each workstream produced. Third, spend, broken down per workstream so managers see what each kind of work costs them.
The reports also throw in recommendations. Anthropic’s own example list includes recurring workflows worth turning into skills and rules worth adding to a code repository. Admins can save custom questions as templates, and anyone with billing permission can raise a teammate’s spend limit straight from the report if a cap is slowing them down.
What the report shows your manager
Picture a bar chart of your team’s month. Click a bar and the sessions inside open up, ranked and filterable by cost, product, date, and output type. Friction shows up too: failed tool calls, missing connectors, the stuff that kept work from finishing.
For a manager, it’s genuinely useful. Claude smart reports show which team leans on AI for real work and which one barely touches it. That’s the pitch Anthropic makes: understand adoption, plan investment, spread what works.
What it does not show (yet)
Here’s where the panic cools down a bit. Anthropic told Mashable that smart reports don’t summarize what workers said in off-topic sessions, and they don’t attach anyone’s name to a flag. The design target is team patterns, not individual dossiers.
Also worth knowing: the AI refuses certain questions entirely. Reports can’t be asked about race, gender identity, religion, political affiliation, immigration status, health, union activity, harassment complaints, or workers’ personal lives. Those topics are blocked at the door.
Now the uncomfortable part. The per-person breakdown is off by default, but it exists. A Mashable follow-up with Anthropic left some ambiguity about how granular individual views can get, and the company’s own support docs describe sample-based transcript analysis. So the honest answer to “can my boss read my Claude chats” is: not word-for-word today, but the structure allows more than team-level summaries, and settings can change.
The off-topic flag, explained
Every report includes what Anthropic calls inefficiencies. Two flavors: sessions that produced nothing usable, and sessions classified as personal or off-topic. Your manager sees how much team time landed in each bucket.
Think about what your last twenty Claude sessions at work looked like. A few pure work. Some mixed, like drafting a report while also fixing your resume because you’re job hunting quietly. One or two entirely personal, like settling a debate about flight prices. The first category builds your reputation. The other two feed the off-topic percentage on a chart someone in leadership will eventually look at.
Anthropic’s spokesperson stressed the flag stays at the team level and says nothing about content. Fair. But percentages travel. A report showing 30% of team time going “off-topic” starts conversations, and those conversations tend to find their way to specific people.
Who can turn this on
Only Enterprise plan admins, for now. Claude smart reports ship with Enterprise plans and live in organization settings, so nothing changes for you personally until someone with admin rights flips the switch. Owners can also delegate reporting to team leads for their own groups, which spreads visibility a little further down.
Smaller plans don’t get the feature at all. If you’re on a Pro subscription you bought yourself, none of this applies. That’s the cleanest privacy guarantee available: nobody can report on what they can’t see.
The money side (because your boss will look)
Cost is half of what makes these dashboards compelling. Each workstream shows spend, so leadership can compare what “vibe coding” costs against what document drafting costs. With the off-topic percentage sitting right next to it, the natural next question is what that slice of budget bought.
One practical angle for your team lead: Anthropic says a capped teammate can get unblocked straight from the report, since admins with billing permission can raise spend limits on the spot. That’s the friendly use case, and honestly the one Anthropic is selling. More budget flowing to the people doing real work is hard to argue with.
Anthropic caps organizations at 10 Claude smart reports per month for free. Additional reports cost extra, which tells you how central usage analytics has become to the Enterprise pitch.
Context matters here. Anthropic filed for IPO in late October, and its filing showed existing customers boosted their spending by 900 percent over the past year, even as the company reported an $80 billion loss in the first half of 2026. Tools that prove AI’s value to a CFO, in charts a CFO can read, are exactly what a company in that position needs to sell. Your off-topic percentage is, from Anthropic’s side of the table, a retention feature.
How to work with it, not against it
You don’t need to quit Claude at work. You need better aim.
- Separate contexts. Work stuff goes in the company Claude. Personal stuff belongs in a free personal account you run on your own time and devices.
- Treat every work session as summarizeable. Before a long chat, ask yourself whether you’d be fine with the topic appearing in a chart with your team’s name on it.
- Put the job-search energy, medical questions, and financial dilemmas somewhere else entirely. They’re exactly the sessions the classifier loves to flag.
- Use Claude for visible wins instead. Draft the spec, build the deck, ship the automation your lead has been asking for. It slots right into Google Docs, Sheets, and Slides if that’s where your work already lives. Reports highlight deliverables, so give it deliverables to highlight.
- If your role involves sensitive material, check with IT about what’s already logged. Enterprise workspaces log and monitor other ways too. Claude smart reports are only one layer on top of that.
Meanwhile, remember that job listings and paperwork aren’t the only risk. Your boss already can’t tell who’s doing the work around here, and a tool that measures AI output both helps and exposes. The people who thrive under monitoring are usually the ones who assumed it was coming.
One more thing worth saying plainly: Anthropic explicitly states smart reports aren’t meant for evaluating individual performance or employment decisions. Surveys find roughly one in four US managers use AI for layoff decisions anyway. Policy and practice don’t always match, so protect yourself accordingly.
Takeaway
Claude smart reports give your employer a team-level window into AI usage, complete with an off-topic flag and per-workstream costs. Keep personal business on personal accounts, make your work sessions count, and don’t type anything into a company tool you’d hate to see summarized in a chart.