I’ll research this thoroughly for you. This is an important transparency question – understanding what your employer can see when you use Claude on an Enterprise plan is crucial for making informed decisions about how you use it.I have the pricing overview. Now let me dig into the actual admin/visibility documentation – this is what really matters for your question.Excellent. I have very strong information. Let me get a few more critical details – specifically about the Compliance API and what data exports actually contain (this is the most important part for your transparency question).This is the key finding. Let me get one more critical piece – the Compliance API specifics and incognito chat behavior.This confirms the most important detail: the Compliance API allows pulling chat data and file content programmatically. Let me check on incognito chats and Claude for Work data privacy specifics.I now have a comprehensive picture. I have everything needed to give you a thorough, transparent answer. Let me put this together as a well-structured guide.
Complete Guide: What Your Employer Can See on Claude Enterprise
This is an important question, and the honest answer is: more than most employees realize. Here’s a complete breakdown so you can make informed decisions.
The Bottom Line First
On the Claude Enterprise plan, your organization’s Primary Owner can access the full content of your chats, uploaded files, and projects — not just metadata. They can do this in three ways: data exports, audit logs (metadata only), and the Compliance API (full content, programmatically). Importantly, incognito chats are NOT hidden from your employer on Enterprise plans.
What Features the Enterprise Plan Includes (Admin Side)
The Enterprise plan includes everything in Team plus the following admin/security capabilities:
Identity & access: SSO, SCIM provisioning, role-based access with fine-grained permissions, domain capture, IP allowlisting, and network-level access control.
Visibility & monitoring: Usage analytics dashboard, audit logs, Compliance API, and the Analytics API.
Data control: Custom data retention controls, organization-wide skills deployment, Google Docs cataloging, admin controls for connectors, and HIPAA-ready offering.
Billing/limits: User-level and org-level spend limits.
By default on Team and Enterprise plans, your content is not used to train Claude’s models — that’s an important protection.
The Three Levels of Visibility Your Employer Has
Level 1: Aggregated Analytics (Always On)
Through the Analytics dashboard and Analytics API, admins automatically see:
- Daily/weekly/monthly active users (WAU/MAU)
- Per-user message counts, conversation counts, projects created, files uploaded, artifacts created
- Which skills and connectors each user uses
- Claude Code metrics (commits, pull requests, lines of code)
- Token consumption per user
- A “leaderboard” of top users by engagement
- Models being used
This is per-user but content-free. Your manager can see that you sent 200 messages last week and used the Slack connector — not what you wrote. Administrators export per-user breakdowns including individual request counts, token consumption, and the primary model being utilized.
Level 2: Audit Logs (Enterprise Only)
Audit logs are available for Enterprise organizations only. Organization Owners and Primary Owners can export logs covering the past 180 days. Importantly, the title and content of chats and projects are NOT available in audit log exports — only their unique identifiers.
So audit logs show events like: “User X created chat ID abc123 at 2:14pm” — but not the chat title or what’s inside it. Audit logs cover sign-ins, session events, file uploads/downloads/deletions, and similar metadata.
Level 3: Full Content Access — Data Exports & Compliance API ⚠️
This is the level most employees don’t know about.
Data Exports: Primary Owners of Team and Enterprise plans can export organization data from Organization settings. Data exports include conversation data and the user data for the account.
Your organization’s designated Primary Owner manages your Work account and all associated data. This includes the ability to request access to your user data through data exports, which may contain your conversations with Claude, uploaded files, and usage patterns.
Compliance API: Enterprise plan Primary Owners can enable the Compliance API. Creating a compliance access key allows pulling activity logs, chat data, and file content programmatically.
This means a Primary Owner can build automated systems that continuously feed every chat and file into the company’s security/compliance tools (Splunk, Datadog, etc.) in real time.
The Incognito Chat Trap — Read This Carefully
Many employees assume incognito = private from the employer. It does not.
If you’re using incognito chats on a Team or Enterprise plan: Incognito chats are included in organizational data exports available to account Owners. While incognito chats aren’t saved to your chat history, they are retained for 30 days, or longer in accordance with your organization’s custom data retention setting.
Incognito chats don’t contribute to memory and aren’t visible in users’ chat histories, but they remain available to Owners through data export features and are subject to your existing data retention policies.
What incognito does do:
- Hides the chat from your own sidebar/history
- Prevents the chat from feeding Claude’s memory feature
- Won’t appear when Claude searches your past conversations
What incognito does NOT do on Enterprise:
- Hide content from your employer’s exports
- Hide content from the Compliance API
- Delete the chat immediately (it’s retained at least 30 days)
Think of incognito as “hidden from me,” not “hidden from my company.”
What This Means for Your Day-to-Day Decisions
Here’s a practical framework for what to consider before typing things into Claude on your work account:
Generally safe to use Claude for at work:
- Any work-related task within your normal job responsibilities
- Drafting work documents, code, emails, analyses
- Research on work-related topics
- Brainstorming work projects
- Learning skills relevant to your role
Think carefully before using Claude at work for:
- Personal matters (resume drafts for new jobs, personal finances, medical questions, relationship issues, legal matters)
- Venting or complaints about colleagues, managers, or the company
- Anything you wouldn’t want surfaced in an HR or legal investigation
- Confidential information about other people (their salaries, performance, personal details)
- Side projects, freelance work, or anything outside scope of employment
- Job hunting, interview prep, negotiation strategies
Never assume privacy for:
- Incognito chats (still visible to admins)
- “Deleted” chats (may already be in exports/Compliance API stream; deletions only affect exports initiated after the deletion)
- Conversations you think are too small to matter (the export captures everything)
Key Principles to Remember
- Treat Claude Enterprise like work email. Same expectation of privacy. Your employer paid for it, owns the account, and has legal/contractual rights to the data.
- The risk is not active surveillance — it’s discoverability. Your manager can’t peek at your chats during lunch. But if your company’s Primary Owner submits an export request, your conversation history and any files you uploaded could be included in that data package. The danger is when something triggers a review (legal hold, investigation, departure, audit), everything becomes accessible.
- Personal accounts exist for a reason. For any non-work use case, use a personal Free, Pro, or Max plan logged into a personal email. Keep work and personal strictly separated.
- Ask your Primary Owner for the actual policy. The use of your Claude for Work plan account is governed by the agreement between Anthropic and your organization. Please contact your Primary Owner for details of this agreement. Companies vary widely — some have permissive personal-use policies, some don’t. Knowing your specific company’s stance matters.
- Custom retention can extend exposure. Default retention is 30 days for incognito and standard org policies, but Enterprise admins can configure custom retention — possibly years. Ask what your company’s retention setting is.
- The Compliance API enables real-time monitoring. If your company has enabled it and integrated with their SIEM, your conversations could be flowing into security monitoring tools as you type, with automated keyword/pattern detection on top.
The Honest Summary
The Claude Enterprise plan is built to give your employer the same level of oversight they have over corporate email, Slack, and Google Workspace — which is significant. Anthropic doesn’t hide this; it’s a selling point to enterprise buyers. The protections you do get on Enterprise are real but pointed at Anthropic (no model training, retention controls, security certifications) — not at your employer.
Use it freely for work. Use a personal account for everything else. That single rule will keep you safe from virtually every scenario where employer visibility could become a problem.
Good question — “audit logs” is a specific technical term and it’s narrower than people often assume. Let me get you the precise list of what’s actually captured.I have the official, definitive list straight from Anthropic’s documentation. Let me lay this out clearly for you.
What Audit Logs Actually Capture on Claude Enterprise
This is the official, complete list from Anthropic’s documentation. Audit logs are an Enterprise-only feature that Owners and Primary Owners can export from Admin settings, covering the past 180 days.
Critically, before going through the events: title and content of chats and projects are not available to be exported in audit logs (only their unique identifiers will be exported). However, chat inputs/outputs will be exportable by Primary Owners via data exports.
So audit logs = metadata only. Content lives in data exports / Compliance API (separate mechanism).
What Each Log Entry Contains
Every audit log row includes these fields:
| Field | What It Reveals |
|---|---|
created_at | Exact timestamp of the event |
actor_info | Who did it (user account info) |
event | The type of event (see lists below) |
event_info | Event-specific details |
entity_info | What was affected (chat, project, file, etc.) |
ip_address | Your IP address |
device_id | Device identifier |
user_agent | Browser/app info |
client_platform | iOS or Android, if mobile |
Translation: every logged event ties an action to you (account), your IP, your device, and the moment it happened.
The Complete List of Logged Events
Authentication & Login Events
user_signed_in_sso— SSO login (records the SSO domain)user_signed_in_google— Google login (records the email used)user_signed_in_apple— Apple login (records the email used)user_signed_out— Sign-outuser_requested_magic_link— Magic link requested (records email)user_attempted_magic_link_verification— Records success/failureuser_sent_phone_code— Phone code sent (records phone number, SMS or call)user_verified_phone_code— Phone code verified
Account Events
user_name_changed— Captures old name and new name
Chat Conversation Events
conversation_created— A new chat was startedconversation_renamed— Records the new name (note: the name of a renamed conversation IS captured, even though chat content isn’t)conversation_deleted— A chat was deleted
Project Events
project_created— New project createdproject_renamed— Project renamedproject_deleted— Project deletedproject_visibility_changed— Records new privacy setting (private/shared)project_document_created— Document added to project knowledge baseproject_document_deleted— Document removed from project knowledge base
File Events
file_uploaded— A file was uploaded
Organization Membership Events
org_user_invite_sent/re_sent/accepted/rejected/deletedorg_user_deleted— User removed from organization
SSO & Security Configuration Events
org_sso_toggled— Records whether SSO is enforcedorg_sso_connection_activated/deactivated/deletedorg_sso_add_initiatedorg_jit_toggled— Just-In-Time provisioning toggledorg_domain_verified/org_domain_add_initiated— Domain capture events
Data Export Events
org_data_export_started— When an export was triggeredorg_data_export_completed— When it finished (and whether Anthropic initiated it)
What Audit Logs CAN Tell Your Employer
From the metadata alone, an admin reviewing logs can determine:
- When you logged in and from where (IP address, device, user agent)
- How often you use Claude and at what times of day
- How many conversations you create, when, and from which device
- When you delete chats (deletions are logged — deleting doesn’t hide that you did something)
- Every file you uploaded (timestamp + file entity ID, though not the filename in the audit log itself)
- Every project you created, renamed, or deleted
- Whether you renamed a chat — and the new name (so don’t rename a chat to something incriminating)
- Authentication anomalies (logins from unusual IPs, magic link attempts, etc.)
What Audit Logs CANNOT Tell Your Employer
- The content of any message you sent or Claude sent back
- The titles of your chats or projects (only UUIDs)
- The content of files you uploaded
- What you talked to Claude about
But — and this is the crucial caveat — everything audit logs hide is visible through the other two mechanisms: Data Exports and the Compliance API. Audit logs are the “lightweight” surveillance layer; the heavy layer is content-level.
Practical Implications
A few non-obvious things worth knowing:
- Deleting a chat is logged. If you delete a conversation right before leaving the company, that deletion event itself sits in the audit log with a timestamp, your IP, and your device. The content may also already be sitting in a prior data export or Compliance API stream.
- Renamed conversation titles are captured in audit logs. Most other metadata is just UUIDs, but
conversation_renamedevents capture the new name inevent_info. So if you rename a chat to something descriptive, that label is in the log. - Logins are heavily tracked. Every sign-in method, IP, and device is captured. Admins can see when you accessed Claude on a personal device versus a corporate one.
- 180-day window for audit logs specifically. Audit log exports cover the past 180 days. But this is separate from the content retention policy your org configures — content can be retained much longer through the Compliance API and data export pipeline.
- Audit log events also flow through the Compliance API. So if your org has the Compliance API enabled and integrated with a SIEM, all of these events stream into security tools in near real-time, not just on manual export.
How This Connects to the Three Layers
To put audit logs in context with the other visibility mechanisms:
| Layer | What It Sees | Who Can Access |
|---|---|---|
| Analytics dashboard / API | Per-user counts, frequency, models, connectors used | Admins & Owners |
| Audit logs | Every action with timestamp, IP, device — but no content | Owners & Primary Owners |
| Data exports / Compliance API | Full chat content, file contents, inputs/outputs | Primary Owner only |
Audit logs are the “who did what when” layer. They’re powerful for security investigations and compliance — and they make clear that even if content were somehow protected, the pattern of your usage is fully visible to the company.
I already pulled this directly from Anthropic’s official audit log documentation in the previous response. Here’s the consolidated answer to exactly what you’re asking:
Audit Logs on Claude Enterprise — What Admins Can Fetch
Audit logs are available only for Enterprise organizations. Organization Owners and Primary Owners can export data within Admin settings > Data and Privacy and clicking the “Export logs” button. Upon triggering this export, all audit logs for the organization within the past 180 days will be aggregated.
Important boundary: title and content of chats and projects are not available to be exported in audit logs (only their unique identifiers will be exported). However, chat inputs/outputs will be exportable by Primary Owners via data exports.
So audit logs = metadata and event tracking only. Actual conversation content lives in a separate mechanism (data exports / Compliance API).
Metadata Captured With Every Single Logged Event
Every single audit log entry includes these fields:
| Field | What It Reveals About You |
|---|---|
created_at | Exact timestamp |
actor_info | Your account info |
event | The action type |
event_info | Event-specific details |
entity_info | Object affected (chat ID, project ID, file ID) |
ip_address | Your IP at the time |
device_id | Your device identifier |
user_agent | Browser/app/OS |
client_platform | iOS or Android (mobile only) |
So even without content, every event is tied to you, your IP, your device, and the exact second it happened.
The Full List of Events That Get Logged
These are the official event types straight from Anthropic’s documentation:
🔐 Authentication Events
user_signed_in_sso— SSO login (logs which domain)user_signed_in_google— Google login (logs email used)user_signed_in_apple— Apple login (logs email used)user_signed_out— Sign-outuser_requested_magic_link— Magic link requested (logs email, success/fail)user_attempted_magic_link_verification— Verification attemptuser_sent_phone_code— Phone code sent (logs number + SMS/call)user_verified_phone_code— Phone code verification
👤 Account Events
user_name_changed— Logs old name → new name
💬 Chat Conversation Events
conversation_created— New chat startedconversation_renamed— ⚠️ Logs the new name in plain textconversation_deleted— Chat deleted (deletion itself is logged)
📂 Project Events
project_createdproject_renamedproject_deletedproject_visibility_changed— Logs new privacy settingproject_document_created— Document added to project knowledge baseproject_document_deleted— Document removed
📎 File Events
file_uploaded— Every file you upload is logged
👥 Organization Membership Events
org_user_invite_sentorg_user_invite_re_sentorg_user_invite_acceptedorg_user_invite_rejectedorg_user_invite_deletedorg_user_deleted— User removed from org
🔒 SSO & Security Configuration Events
org_sso_toggled— Logs if SSO is enforcedorg_sso_connection_activatedorg_sso_connection_deactivatedorg_sso_connection_deletedorg_sso_add_initiatedorg_jit_toggled— Just-In-Time provisioning toggleorg_domain_verifiedorg_domain_add_initiated
📤 Data Export Events
org_data_export_started— Logs whether Anthropic or admin started itorg_data_export_completed
Quick Reference: Visible vs. Hidden in Audit Logs
| ✅ Admins CAN see from audit logs | ❌ Admins CANNOT see from audit logs |
|---|---|
| When you logged in and from where (IP) | The content of your messages |
| Which device/browser you used | Claude’s responses to you |
| How often you use Claude | Chat titles (only UUIDs) |
| Every chat created or deleted | Project titles (only UUIDs) |
| Every file uploaded (with timestamp) | File contents |
| Every project action | File names (only UUIDs) |
| Renamed conversation labels ⚠️ | What you talked about |
| Project visibility changes | |
| Authentication patterns | |
| Login anomalies |
Three Things Worth Highlighting
1. Renamed conversations leak text into audit logs. Most chat metadata is hidden behind UUIDs, but the conversation_renamed event captures the new name. If you (or Claude) rename a chat to something descriptive, that text is in the audit log even though chat content isn’t.
2. Deletion is logged. Deleting a chat doesn’t erase your tracks — the deletion event itself is recorded with your IP, device, and timestamp.
3. Audit logs are content-free, but they’re not the only mechanism. Audit logs alone won’t expose what you wrote. But on Enterprise, Primary Owners can also pull full conversation content via data exports and the Compliance API. Audit logs are just the lightweight surveillance layer; full content access exists separately.
If you want, I can break down what’s in data exports specifically (the layer that actually contains your conversation content) — that’s the bigger privacy lever to understand.
Yes. Short answer: your Enterprise Primary Owner can read your code, your prompts, and Claude’s responses from Claude Code sessions. Let me explain clearly why, because I think the confusion is around how command-line usage connects to your company’s admin visibility.
How Claude Code Actually Works (The Key Insight)
When you run Claude Code in your terminal, it feels local — you’re typing in your own command line, on your own laptop, in your own repo. But here’s what’s actually happening under the hood:
Your terminal → [code + prompts sent over internet] → Anthropic's servers
↓
Claude processes it
↓
Your terminal ← [response sent back] ← Anthropic's servers
Claude Code is not a local AI model. The Claude model lives on Anthropic’s servers. So every time you ask Claude Code to do something, your code and prompts leave your machine and travel to Anthropic to be processed.
The terminal is just the interface. The actual work happens on Anthropic’s cloud — under your company’s Enterprise account that you’re authenticated against.
What Specifically Gets Sent to Anthropic’s Servers
Every Claude Code session transmits:
| What you do locally | What gets sent to Anthropic |
|---|---|
| You type a prompt | The full prompt text |
Claude reads auth.py | The complete contents of auth.py |
| Claude reads 20 files to understand context | All 20 files’ contents |
Claude runs git log | The git log output |
Claude runs cat .env | Your environment variables (yes, including secrets if exposed) |
| Claude proposes a code edit | The full diff |
| Claude runs tests | The test output |
| You have a 2-hour debugging session | The entire session transcript |
All of it sits on Anthropic’s infrastructure, tagged to your user account inside your company’s Enterprise organization.
Why Your Enterprise Owner Can Access It
Because your account belongs to the company’s Enterprise organization, everything your account does flows into your company’s data scope. The Primary Owner has two mechanisms to retrieve it:
1. Data Export — Primary Owner clicks a button, gets a download containing conversation data and uploaded files for the org, including Claude Code sessions.
2. Compliance API — Primary Owner enables programmatic, real-time access. The Compliance API explicitly provides real-time, automated access to Claude Code’s usage data and conversation logs. Companies often pipe this directly into security tools (Splunk, Datadog) so it’s monitored continuously, not just on demand.
So when you run Claude Code, you should mentally picture:
“I am sending this code and these prompts to my company, with a copy going through Anthropic, and they can both review it whenever they want.”
That’s not paranoia — that’s literally the architecture.
The Common Misconception
Many developers assume: “It’s my terminal, it’s my laptop, it’s my local repo — so this is private.”
That intuition is wrong for any cloud-based AI coding agent (Claude Code, Cursor, Copilot, Codex, Cody — all of them work this way). The local interface is just a window into a cloud service. The moment Claude Code reads a file, that file’s contents are on someone else’s server.
The only AI coding tools that are actually private from cloud providers are fully local models (running on your own GPU, like Ollama with a local LLM) — and those aren’t what Claude Code is.
What This Means for Your Daily Use
For company repos and company work: Use Claude Code freely. Your company already owns the code, and they’re paying for visibility into how their AI tools are used. This is the intended use case.
For anything else, do not use your company Claude Code account:
- ❌ Don’t
cdinto your personal side project and run Claude Code — your personal code is now in your company’s Enterprise data - ❌ Don’t use it on a freelance client’s repo — you’ve leaked their IP into your employer’s audit trail
- ❌ Don’t use it on code from a previous employer — same problem
- ❌ Don’t paste API keys, passwords, or secrets into prompts — they’re now permanently logged
- ❌ Don’t use it to prepare for job interviews (e.g., “help me solve this LeetCode problem for my Google interview tomorrow”) — visible to your current employer
- ❌ Don’t run it in directories with personal files (tax documents, journals, medical info) — Claude might read those files
The clean rule: When you launch Claude Code, you should only be in directories containing code your employer is allowed to see. If you wouldn’t git push it to the company GitHub, don’t open Claude Code in that folder.
The Right Mental Model
Treat Claude Code on your Enterprise account exactly like:
A company-owned laptop with screen recording enabled.
Everything you do is technically usable by you, and most of it will never be reviewed — but all of it is retrievable if your company ever needs to look. That single mental model will keep you safe.
For personal coding, side projects, learning, and anything outside work scope: get a personal Claude Pro or Max subscription on a personal email. That’s a separate account, separate data scope, and your employer cannot see anything in it.