Skip to main content
    Back to Blog
    12 min read
    Workplace Privacy
    AI Privacy
    ChatGPT
    Claude
    Gemini
    Microsoft Copilot

    Can My Employer See My AI Chats? ChatGPT, Claude, Gemini and Copilot

    T

    Team Sonomos · Sonomos

    Team Sonomos writes about AI, privacy, and what happens to sensitive information at work.

    Your employer can see your AI chats when it controls any one of three things: the account you are signed into, the computer you are using, or the network your traffic crosses. Every major vendor documents a way into work accounts: the Compliance API on Claude Enterprise, Google Vault for Gemini in Google Workspace, Microsoft Purview for Copilot, and admin export and compliance tools on ChatGPT Business and Enterprise. A personal account, used on a personal device over a personal network, gives your employer no route in through any of them. And deleting a chat seldom settles the matter, because workplace retention rules and legal holds keep copies the app no longer shows you.

    This page compares what each vendor's own documentation says as of November 18, 2026, and links to our detailed guide for each tool. These products change monthly, and two of the changes below took effect in November, so check the date on any answer you rely on.

    What admins can see, export and keep

    The table reads across: which account you use, whether your employer can read what you wrote, the tool it would use, and what happens when you delete a chat.

    AccountCan your employer read the chats?HowIf you delete a chatOur guide
    ChatGPT Free, Plus or ProNoNo admin access through OpenAIYour own settings decideChatGPT
    ChatGPT BusinessYesWorkspace admins can access and export conversations, according to OpenAIRemoved within 30 days; Business has no custom retention settingChatGPT
    ChatGPT Enterprise or EduYesAdmin console and Compliance API; a workspace owner grants the "Conversation messages" permissionAdmins set retentionChatGPT
    Claude Free, Pro, Max or UltraNoNo admin access through AnthropicYour own settings decideClaude
    Claude TeamYes, through an exportPrimary Owner's organization data export; Cowork activity events, with prompt text only if IT enables itLeft out of exports made after the deletionClaude
    Claude Enterprise, chatYesCompliance API; audit logs exclude titles and contentContent removed, but the chat stays listed with a deletion dateClaude
    Claude Enterprise, CoworkYesCompliance API, including sessions run on your computer, kept six years by defaultGoverned by retention, not by your deletionCowork
    Gemini, personal Google accountNoNo admin access through GoogleYour own settings decideGemini
    Gemini, Workspace accountYes, on editions with VaultGoogle Vault search and export; Admin console log eventsVault holds and retention rules override itGemini
    Copilot, personal Microsoft accountNot through MicrosoftOn a managed PC, Purview can capture prompts typed in EdgeYour choice; Microsoft's pages for the current app state no retention periodCopilot
    Copilot, work (Entra ID) accountYesStored in your mailbox: eDiscovery, retention, audit, DSPM for AIStays searchable until retention ends; holds stop deletionCopilot

    Sources for the table: Anthropic's Compliance API reference, its organization data export article (updated November 10, 2026) and its Cowork article for Team and Enterprise (updated November 16, 2026); Google's Vault guide to the Gemini app; Microsoft's Copilot privacy page (updated November 3, 2026), Copilot retention page and privacy pages for the consumer Copilot app released on August 18, 2026. The ChatGPT rows reflect OpenAI's documentation as reviewed for our ChatGPT guide on November 2, 2026, and OpenAI's plan comparison, which lists custom retention controls only for Enterprise and Edu.

    Two patterns stand out. Work accounts are reachable on every platform, and the differences are about method and plan: Claude Team gives owners an export and usage data rather than a search tool, and Gemini's Vault coverage depends on the Workspace edition. And the personal-account rows all say no for the account itself, which is a narrower promise than it sounds, for the reasons in the next section.

    Three things decide it: the account, the computer, the network

    The account

    If you signed in through your company's single sign-on, with your work Google account, or with the Microsoft account your employer manages, the account belongs to your employer, whatever the app looks like. Anthropic tells Team and Enterprise members in plain terms that the organization's Primary Owner manages your work account and all associated data, including the ability to request your data "through data exports, which may contain your conversations with Claude, uploaded files, and usage patterns." Google's Vault lets authorized staff search Gemini app conversations by account or organizational unit and export the results. Microsoft stores Copilot conversations in a hidden folder of the work mailbox, where compliance staff search them with eDiscovery.

    Microsoft has made the account harder to spot. Its Microsoft 365 app is now the Microsoft Copilot app, serving work and personal accounts in one place, and Microsoft says the app shows "clearer visual indicators" of which account you are in. Check that indicator before you type anything you would not put in a work email.

    The computer

    A work computer can capture what you type into any AI site, on any account. Microsoft documents this most fully: its Purview page on other AI apps describes capturing prompts typed into ChatGPT, Google Gemini, consumer Copilot and DeepSeek in the Edge browser, using a browser extension on devices onboarded to Purview and a collection policy set to capture content. Endpoint data loss prevention on onboarded Windows computers can warn or block when someone pastes sensitive data into an AI site.

    Anthropic has added its own controls for managed machines. A beta setting announced on November 5, 2026 lets Enterprise owners make Claude Desktop refuse personal accounts on work computers. And when Claude Cowork runs locally, its history is a file on that computer. Incognito mode changes none of this, because these tools sit below the browser tab or inside it.

    The network

    Traffic to AI services is encrypted, so an office network or VPN on its own sees where you went and when, not what you typed. The exception is a security gateway that inspects encrypted traffic: Microsoft documents a "network data security" integration that analyzes AI prompts and responses "at the network layer" through a secure access service edge product. Inspecting encrypted traffic generally needs the device to trust the company's gateway, which managed devices do and personal ones usually do not. On your own phone or laptop, the network sees the destination and the timing.

    Deleting a chat rarely deletes your employer's copy

    Each vendor separates what you see from what the organization keeps, and each says so in its own words.

    Google's June 11, 2026 announcement of Vault retention rules and holds for the Gemini app gives the clearest example: "If a user deletes a conversation or turns off their activity setting, but an active Vault hold requires retention, the data is hidden from the user but remains fully retained and visible to Vault administrators."

    Microsoft's retention page says that where a retention policy applies, deleted Copilot messages move to a hidden folder where they stay searchable with eDiscovery until the retention period runs out, and that a hold suspends permanent deletion. It warns: "Messages visible in your AI apps are not an accurate reflection of whether they are retained or permanently deleted for compliance requirements."

    Anthropic's Compliance API reference says that when a Claude Enterprise user deletes a chat, its content is deleted, but the API "still lists the chat, with deleted_at populated and an empty name." For Cowork sessions that ran on your computer, the session transcript documentation says visibility "is governed by retention instead," six years by default. On Team plans, a deleted chat is left out of any organization export made afterwards, but not out of one already made.

    Anything an endpoint tool captured as you typed, or a security stream already exported, sits on the company's own systems. Our guide to AI chats as evidence explains why a company facing a dispute may be obliged to keep it.

    What changed in 2026

    These are the dated changes behind the table, oldest first.

    • April 9, 2026. Claude Cowork becomes generally available on macOS and Windows, with usage analytics and OpenTelemetry monitoring for Team and Enterprise admins, according to Anthropic's release notes.
    • June 11, 2026. Google Vault adds retention rules and litigation holds for the Gemini app, on top of the search and export it already offered.
    • November 5, 2026. Anthropic makes inference hooks generally available for Claude Enterprise, sending prompts from supported Claude products to a company's security tool for a check, and adds a beta setting that blocks personal accounts in Claude Desktop.
    • November 16, 2026. The unified Chat and Cowork experience turns on by default for Claude Enterprise organizations that had not chosen.
    • December 1, 2026. It turns on by default for Claude Team, unless an owner keeps it off.
    • February 1, 2027. Claude Team's separate chat and Cowork experience is retired.

    In the United States, employers generally may monitor the accounts and devices they provide, especially where they have told employees they do. The federal wiretap statute allows interception where "one of the parties to the communication has given prior consent" (18 U.S.C. § 2511(2)(d)), and employers typically collect that consent through the acceptable use policy you agree to. Some states add notice rules: New York's Civil Rights Law § 52-c requires employers that monitor email, internet access or usage to give prior written notice on hiring, get an acknowledgment, and post the notice where employees can see it.

    In Europe, monitoring has to meet data protection rules, and Article 88 of the GDPR lets member states set specific rules for employee data, with safeguards that pay particular attention to "monitoring systems at the work place." None of this is legal advice, and the answer for your job depends on your country, your contract and your employer's policies. Using an AI tool your employer has not approved can be a policy problem in its own right, which is the shadow AI issue. If you are applying for a job rather than holding one, the picture is different again: see whether an interviewer can see your AI history.

    Frequently asked questions

    Can my boss see my AI chats if I use a personal account at work?

    Not through the AI vendor, since your employer does not administer a personal account. But a work computer can capture what you type into any AI site, and Microsoft documents how Purview does this for ChatGPT, Gemini and consumer Copilot in Edge. A personal account is private from your employer only on a personal device and a personal network.

    Which AI chatbot keeps my chats most private from my employer?

    On a work account, none of them. ChatGPT, Claude, Gemini and Copilot all document admin tools for searching, exporting or retaining work conversations. The account type matters far more than the vendor: a personal account on a personal device is the private option with every one of them.

    Can my company see AI chats I have already deleted?

    Often, yes. Google Vault holds override user deletion for the Gemini app, Microsoft keeps deleted Copilot messages searchable until retention ends, and Claude Enterprise keeps a record that a deleted chat existed. Copies captured by endpoint tools or exported earlier are not affected by deletion at all.

    In the United States it generally is on employer-provided accounts and devices where you have been told about monitoring, and some states, such as New York, require written notice. European law adds limits on workplace monitoring. Check your contract and acceptable use policy, and take legal advice for a specific dispute.

    Before you type, check three things

    Ask whose account it is, whose computer it is and whose network it is. If the answer to any of the three is your employer, write as you would in a work email, because the same retention, search and export tools now reach AI chats too. Keep personal questions on a personal account and a personal device, and keep company data inside the tools your employer has approved. Our guide to personal AI privacy covers the settings worth changing on your own accounts.

    The one thing fully in your hands is what goes into the prompt. Locke, made by Sonomos, runs on your own computer and replaces names, account numbers, health details and other sensitive values with placeholders before a prompt reaches the AI tools it screens, so the provider never receives them. It does not hide AI use from your employer or make a work account private, and monitoring software on a work computer can still record what you type. Locke is free for one person on x64 Windows and Linux, and you can download it here.

    Keep personal details out of what the AI receives

    Locke Free runs on your own computer. Before a prompt reaches the AI tools it screens, it replaces the names, account numbers, health details and other sensitive values it detects with placeholders, so the AI provider never receives them. Neither does Sonomos, which makes Locke and publishes this blog.

    It does not hide your AI use from your employer or make a work account private. Admin tools, retention and device monitoring work as this article describes, and monitoring software on a work computer can record what you type before Locke masks it. Detection can miss things, and coverage depends on the app and platform.

    Free for one person on x64 Windows and Linux. macOS is coming soon.

    This article is general information about privacy and AI, not legal advice, and reading it does not create a solicitor–client or attorney–client relationship. Regulatory requirements turn on your specific facts and jurisdiction and change over time. Consult qualified counsel before relying on anything here, and check the publication and last-reviewed dates above. We may revise posts, but we do not undertake to keep any post current as the law changes.

    Reference pages for this post