Claude Can Now Write Your Boss a Report on How You Use It. Here’s What It Can and Can’t See.

If your company pays for Claude, your manager can now ask Claude to read a sample of your team’s conversations and write up what everyone has been doing with it. Anthropic made the feature, called smart reports, generally available to Claude Enterprise customers on October 28, after a beta that started in September. UNILAD Tech’s headline put it bluntly: a feature that “lets bosses read employees’ AI chats.”

That’s partly right and partly not. Smart reports don’t hand managers your transcripts. They hand managers an AI-written summary of transcripts, with names hidden unless the company turns attribution on. The bigger surprise for many workers may be what was already possible before this launched. Here’s what the feature does, where Anthropic drew lines, and what you should assume about any AI chat you have at work.

What a smart report actually contains

According to Anthropic’s help center, smart reports “analyze how a team uses Claude and report on the work getting done, what it costs at list price, where sessions run into friction, and which repeated patterns are worth packaging as shared skills.” An admin chooses a team, a date range of up to the last 28 days, and which products to include: regular chat, Claude Code, Claude Cowork or all of them. They can add focus areas and custom questions. Claude then reads a sample of the transcripts and writes the report, which usually takes a few hours and downloads as a web page.

The sections read like a consultant’s slide deck: workstreams, deliverables, cost per session by type of output, task outcomes, pain points, inefficiencies, complex autonomous work, the most expensive sessions, and recommendations. In its launch post, Anthropic gives “friction” examples like a missing connector or a tool call that failed. Customers named in the post include Wix, Deloitte, DoorDash, Docusign and Monday. Wix said each report “is based on tens of thousands of sessions.”

Each organization can run 10 reports a month at no charge, with queued and in-progress reports counting toward the cap. Owners can also delegate report access to team leads for up to 50 groups or departments.

Where Anthropic drew the lines

Anthropic built in a few guardrails that admins can’t switch off. Custom questions can’t target restricted topics. The restricted list covers protected characteristics, health and medical matters, union activity, harassment and whistleblower issues, political and religious views, personal life, “feelings about a specific named person,” and privileged legal content about the workplace. Sessions on those topics “are used only for aggregate analysis and don’t include summaries,” and “personal conversations also don’t include individual sessions or names.”

Names are the main dial companies control. A setting called “Allow attribution to individual users” is off by default, which means people appear as “User” and session IDs are hidden. If an Owner turns it on, report viewers can reveal names, and Anthropic says “unveils are logged.” The company is also explicit about intent: reports “aren’t designed and should not be used for evaluating individual performance or making employment decisions.” That’s a policy statement, not a technical block. Nothing stops a manager from reading a summary and drawing conclusions.

Smart reports aren’t available to companies using customer-managed encryption keys, a HIPAA configuration, or a hybrid setup that runs the AI on their own cloud. If Claude Code runs with zero data retention, reports cover only chat and Cowork.

Your boss could already see more than this

Here’s the part that gets lost in the headlines. On Team and Enterprise plans, the Primary Owner can already export the organization’s data, including conversations. On Enterprise, the Compliance API lets a company pull “activity feed events, chat data, and file content” into its security and legal tools. Incognito chats on work plans are kept for at least 30 days and show up in both. And unless a company sets a custom retention period, Enterprise data “is retained indefinitely.”

Rivals work the same way. OpenAI’s Compliance Platform pipes ChatGPT Enterprise conversation logs into eDiscovery and data-loss tools. Microsoft records Copilot prompts and responses in its Purview audit log, though its Copilot Dashboard shows only group-level stats with a minimum group size of five. Google Vault started retaining and exporting Gemini app conversations in June 2026, even ones users delete. What’s new with smart reports isn’t access to the text. It’s an AI that reads the text for your manager and tells them what it means, which is a different kind of visibility. Microsoft is heading the same way from the security side, too: it now treats AI agents like insider threats and sells tools to watch them.

Is this really “restricted in over 30 countries”?

UNILAD’s headline says the practice is restricted in more than 30 countries. We couldn’t find a primary source for that exact count, and Anthropic’s documentation doesn’t mention countries at all. The figure likely comes from guides that track employee monitoring laws across 30-plus countries. What’s true is that most developed countries regulate workplace monitoring, and most of those rules require notice, a legitimate reason or a say for workers, rather than banning it outright.

• Europe: In the 2017 case Bărbulescu v. Romania, the European Court of Human Rights ruled that monitoring an employee’s work messages violated their privacy because they hadn’t been clearly warned. Prior notice and proportionality are now the baseline.
• Germany: Works councils have a legal right to co-decide on any technology “suitable” for monitoring staff behavior or performance. A tool like smart reports would almost certainly need their sign-off.
• EU AI Act: AI that infers employees’ emotions at work has been banned since February 2025. Smart reports analyze tasks, not feelings, so that ban likely doesn’t apply, but the stricter rules for high-risk workplace AI arrive in December 2027.
• New York, Connecticut and Delaware require employers to give written notice before monitoring electronic communications. New York’s law, in force since May 2022, carries fines of $500 to $3,000 per violation.
• California gives employees the right to know, access, correct and delete personal data held about them, including inferences, since the state’s privacy law began covering workers in January 2023.

Monitoring is already normal at American workplaces. In an ExpressVPN survey of 1,500 US employers published in February 2025, 74% said they use online monitoring tools, and 61% said they use AI to analyze employee performance. Meanwhile, most knowledge workers were already using AI by 2024, and many brought their own tools, according to Microsoft and LinkedIn’s Work Trend Index.

What the backlash gets right

The reaction UNILAD collected from Reddit captured both sides. One user wrote: “On one hand, I hate the idea of my manager seeing my chat logs. On the other hand, it’s pretty realistic that lots of engineers will use their work tokens on side projects/businesses if there’s no consequences, which would be a huge money suck on the company considering the price of frontier AI right now.” Another was shorter: “Claude needs to learn that snitches get stitches.”

The cost point is real. Anthropic’s own pitch for smart reports leans on spend: what each kind of work costs at list price and which sessions are the most expensive. With AI bills now a serious line item, as Anthropic’s own S-1 filing makes clear from the other side, finance teams want to know what they’re paying for. The privacy worry is real too. People talk to chatbots differently than they write emails. They think out loud, admit what they don’t know and ask the questions they’d never ask a colleague. A tool that summarizes those conversations for management changes how candid anyone will be, even with names hidden.

Smart reports are probably not the end of this trend. Every enterprise AI vendor is under pressure to prove its product pays for itself, and the easiest proof is a report that shows managers exactly what their staff are doing with it. The difference between a useful tool and a surveillance tool will come down to settings most employees never see, which is why it’s worth asking about them now.

Sources and further reading

• UNILAD Tech: New Claude feature lets bosses read employees’ AI chats
• Claude Help Center: Get started with smart reports
• Claude Help Center: Let team members run smart reports
• Anthropic: Smart reports, surveys and pooled group budgets for Claude Enterprise
• Claude Help Center: Release notes
• Claude Help Center: Export your organization’s data
• Claude Help Center: Access the Compliance API
• Claude Help Center: Use incognito chats
• OpenAI: Compliance Platform for Enterprise
• Microsoft Learn: Purview and Microsoft 365 Copilot
• Microsoft Learn: Copilot Dashboard settings
• Google Workspace Updates: Vault retention for the Gemini app
• Columbia Global Freedom of Expression: Bărbulescu v. Romania
• Future of Privacy Forum: The EU AI Act’s workplace emotion recognition ban
• Morrison Foerster: New York’s employee monitoring notification law
• Morgan Lewis: California privacy law now covers employees
• ExpressVPN: Workplace surveillance survey
• Microsoft and LinkedIn: 2024 Work Trend Index

About this article: GeekBlog covers U.S. technology news, AI, phones, smartwatches and gaming. Every story is written and checked under our Editorial Policy. Spotted a mistake or have a story tip? Contact our editors.