Can Anthropic See My Claude Chats?

Not by default. On Free, Pro and Max, Anthropic’s privacy page says its staff can’t see your Claude chats unless you send one as feedback (the thumbs buttons) or a Usage Policy review needs it. On Team and Enterprise, the help pages don’t repeat that rule, but the Commercial Terms let Anthropic share your content only with employees who need to know it. Deleted chats still sit on Anthropic’s back end until they’re purged, within 30 days, incognito ones as long (longer on Enterprise with a long retention period), and flagged ones far longer. So client secrets go only into a company seat, where Anthropic is your processor and doesn’t train on chats by default, and nobody rates a chat that holds them. Here’s what to tell staff.
I read Anthropic’s privacy, retention, training and usage pages and its Commercial Terms on November 18, 2026. None of it was tried out in a live workspace for this post, and it isn’t legal advice.
We ran on personal Pro and Max subscriptions, then moved our company to Claude Team in mid-2026, and I’m the Primary Owner of that org. So the owner’s checklist below is written from the seat that has the switches.
Does Anthropic read my chats?
Software checks them, and people see them only in a few named cases. Anthropic’s Usage Policy says its Safeguards Team “implements detection and monitoring,” without saying how. The flag itself comes from what the retention pages call “automated trust and safety systems”: classifiers, not staff.
For people, the clearest rule is on Anthropic’s consumer privacy page: “By default, Anthropic employees cannot access your conversations unless” you consent by giving feedback, or “Review is needed to enforce our Usage Policy.” Then “only designated members of our Trust & Safety team may access this data on a need-to-know basis.” That page covers Free, Pro and Max.
Two more groups exist on personal plans. With the training setting on, Anthropic says it removes your user ID from chats before review: “Access is limited to a small number of personnel involved in model training.” And the privacy policy says that for flagged content, “we may re-identify the Inputs or Outputs to enforce our Terms of Service or Usage Policy.” So the de-linking covers training only, and enforcement can tie a flagged chat back to you.
On Team and Enterprise, the consumer privacy policy doesn’t apply. Anthropic is a processor that “only processes the data as instructed by the customer.” I found no Team or Enterprise help page that repeats the staff-access rule. The Commercial Terms cover it in general words: “Customer Content is Customer’s Confidential Information,” and Anthropic may share it only with “employees, agents, and advisors that have a need to know.” They don’t say which team that is or what triggers a look. The one commercial help page with a similar sentence, about covered models, applies only to organizations on zero data retention.
What gets a chat flagged in a normal firm’s work?
Anthropic publishes what’s prohibited, not what its classifier looks for. The closest thing to ordinary office work in the Usage Policy is security: it bans trying to “Discover or exploit vulnerabilities in systems, networks, or applications without authorization of the system owner,” and allows testing on systems you own or are authorized to test. If your IT person pastes a scan of a client’s network into Claude, have the client’s written OK first.
Legal, medical, tax, investment and insurance recommendations or decisions about a person are “high-risk use cases,” where a qualified person must stay accountable. Internal drafting, research and summaries that aren’t the final recommendation are excluded. That’s a rule about how you use the answer, and nothing says it trips a flag. The policy also says hitting a block “does not by itself indicate a violation.”
A flag matters because it outranks deletion and incognito. Anthropic keeps a flagged chat’s inputs and outputs for up to 2 years, and the trust and safety classification scores (the classifier’s verdict, not the chat) for up to 7 years.
Does clicking thumbs up or down send my chat to Anthropic?
Yes, the whole conversation. Anthropic’s training page for company plans says a thumbs rating stores “the entire related conversation, including any content, custom styles, conversation preferences, or model settings, in our secured back-end for up to 5 years.” Bug reports count too: the retention page keeps “data associated with that submission” for 5 years.
This is the everyday way a chat leaves your control. It works on Team and Enterprise, where it’s one of the exceptions to no training, and it ignores the training setting on personal plans. A staff member who clicks thumbs-down on a bad answer about a client’s tax return has just sent that return to Anthropic’s feedback store. A Primary Owner or Owner can turn off the Rate chats setting, which stops everyone in the org from sending ratings; that’s step 1 below. The training post covers the other ways a Team org opts in.
Does Claude store deleted chats?
For a while. On every plan, a deleted chat is “Removed from your chat history immediately” and gone from Anthropic’s back-end storage within 30 days. The company-plan page uses the same words. “Within” is a ceiling, so a deleted chat may go sooner.
Three main things outlast the purge: a flag (above), a thumbs rating (above), and the law. “In all cases, we may retain your chats or sessions as required by law,” the company page says, and the personal page adds “to resolve disputes.” On a personal plan with training on, deleting a chat keeps it out of future training, but “Your data will still be included in model training runs that are already in progress, or in models that have been trained.” Both retention pages also let Anthropic keep anonymized data longer for research; on company plans, only “If permitted in your contract with us.”
Deletion on Anthropic’s side is half the job. The client-data post lists the places on your side that a deleted chat leaves behind.
How long does Claude store deleted chats?
Up to the ceiling above, unless one of the three exceptions applies. The chart puts every clock on one scale.
Show the data
| Case | How long Anthropic can keep it | Plans | Source |
|---|---|---|---|
| Deleted chat | within 30 days after you delete it | Any plan | Retention, company plans and personal plans |
| Incognito chat | within 30 days, or longer under an Enterprise org’s custom period | Any plan | Retention, company plans, incognito guide |
| Shortest custom period an Enterprise owner can set | 30 days; with nothing set, kept indefinitely | Enterprise | Custom retention |
| A chat nobody deletes | Until someone deletes it | Team (and Enterprise with nothing set) | Retention, company plans |
| Flagged chat: inputs and outputs | up to 2 years | Any plan | Both retention pages |
| Training setting on, de-identified | up to 5 years | Free, Pro, Max | Retention, personal plans |
| Thumbs rating: the whole conversation | up to 5 years | Any plan | Training, company plans |
| Flagged chat: trust and safety classification scores | up to 7 years | Any plan | Both retention pages |
Can Anthropic see incognito chats?
Incognito keeps a chat out of your history and out of Claude’s memory. It doesn’t keep it out of Anthropic’s back end. The company retention page says incognito chats are “automatically deleted within 30 days unless flagged as a Usage Policy violation.” The incognito guide says they’re kept “for either 30 days (default), or longer in accordance with your organization’s custom data retention setting,” which only Enterprise has. An Enterprise org with a long retention period keeps incognito chats that long too.
What incognito does promise: incognito chats are never used to train Claude, even with the training setting on.
On Team and Enterprise, the same guide says incognito chats “are included in organizational data exports available to account Owners,” and on Enterprise in the Compliance API. So incognito hides a chat from your history, and your Primary Owner can still export it. The employer post covers that side, and the bank-statements post what it means for an uploaded file.
Does Claude train on my chats, and how long does that keep them?
On personal plans it’s your choice, and the choice changes how long Anthropic keeps your data. Since the consumer terms change announced August 28, 2025 (existing users had until October 8, 2025 to decide), training is a setting. With it on, Anthropic keeps data “in a de-identified format for up to 5 years in our model training pipelines”, for new or resumed chats only.
Company plans are different: by default, Anthropic doesn’t train on Team, Enterprise or API chats. The exceptions are feedback and the Development Partner Program, which you’d have to join. Does Claude Team train on your data? has the full answer.
When a buyer of mine is on Claude Pro, we turn the training setting off together at the setup check, and our own deployments run on Claude Team. Training off still leaves you on consumer terms, which is why client work belongs on a company plan.
How is Claude data retention different on Team and Enterprise?
Team keeps chats “in the product” until someone deletes them, and has no custom retention. Enterprise lets a Primary Owner or Owner set a retention period of at least 30 days, counted from a chat’s last activity. Two catches: with nothing set, “data is retained indefinitely,” and chats inside a project follow the project’s period, which is also indefinite by default.
That minimum is the shortest period an owner can pick, not how long Anthropic keeps a chat. The flagged and legal rules apply on every plan. The training post’s Enterprise section covers what custom retention doesn’t reach.
Does Claude keep sensitive data private?
From other users, yes. From Anthropic, only as far as the rules above go, and an uploaded file follows the same rules as typed text. Anthropic’s training pages say training and feedback data don’t include raw content from connectors (Google Drive, MCP servers) unless it’s copied into the chat. I found nothing on how long Anthropic keeps what a connector reads into a chat, so treat it like anything you typed.
Anthropic’s own advice on personal plans is to be thoughtful about “Confidential business or personal documents,” health records, passwords and account numbers. For a firm, the legal line matters more: Anthropic “may retain” chats “as required by law,” and under the Commercial Terms, Anthropic must tell the customer promptly about a compelled disclosure unless that’s prohibited. How that plays out for a deleted chat in litigation is in the privilege post. Claude Code sends code to Anthropic on its own retention clock, covered in the Claude Code post.
When a client asks what contract covers their data: on Team and Enterprise, the Commercial Terms say data “will be processed in accordance with the Anthropic Data Processing Addendum,” so the DPA comes with the plan. Health data is different, because Team can’t enable HIPAA; is Claude safe for business has what Enterprise adds. A tool your firm builds on the API runs on the API’s clock: Anthropic deletes API inputs and outputs within 30 days, unless you’ve agreed otherwise, as with a zero data retention agreement.
The cheapest privacy is not sending the data at all. On one client build, we planned pre-launch tests on made-up projects and messages, to keep the client’s data private while we built. In our setups, connectors run inside the client’s own account, and we keep no copy of their data.
Who can see my Claude chats: Anthropic, my employer or my teammates?
Anthropic’s staff and its classifiers are two of five audiences, and each has its own door.
| Who | A normal chat | An incognito chat | A deleted chat |
|---|---|---|---|
| Anthropic staff | Personal plans: if you rate it, a Usage Policy review needs it, or (training on) training staff review it de-linked from you. Team and Enterprise: employees with a need to know, under the Commercial Terms | Same rules, until it’s purged | Same rules, until it’s purged; a rated or flagged one longer |
| Anthropic’s classifiers | Yes, they check chats for Usage Policy violations | Yes, and a flag keeps it past the purge | A flag made before you deleted it outlasts the deletion |
| Your Primary Owner (Team, Enterprise) | Yes, in an org export | Yes, in an org export | No, not in exports made after you delete it |
| Enterprise admins, through the Compliance API | Yes, if it’s set up | Yes | Listed, without its content |
| Teammates | Not unless you share it | No page says | No; deleting a chat removes a shared copy too |
Your employer’s side is in can my employer see my Claude chats, and coworkers’ in can teammates see my Claude chats. If Claude acts in your browser, is Claude in Chrome safe covers what it can reach there.
What should I tell staff about Anthropic and their chats?
Change the settings first, then send the note. Steps 1 and 3 need the Primary Owner or an Owner.
- Turn off thumbs ratings (Team and Enterprise).
- Open Organization settings, then Data and Privacy.
- Find Rate chats.
- Turn it off, unless your chats never hold client data. The Team setup post walks through the rest of that page.
- Check the training setting on every personal seat someone uses for work.
- Click your name, then Settings.
- Open Privacy.
- Under “Help Improve our AI models,” switch it off.
- Then move client work to a company seat anyway.
- On Enterprise, set a retention period.
- Open Organization settings, then Data and Privacy.
- Set the period (at least 30 days).
- Save.
- Set a period on client projects too, or their chats stay indefinitely.
- On Team, have staff delete finished client chats. Team has no retention setting, so a chat stays until someone deletes it. Put a monthly reminder on the calendar, and decide what happens to a departing member’s chats (the offboarding post covers it).
- Write down what your vendor keeps. Our company’s rule is that every paid AI provider we call gets a page in our records noting its data-retention terms. A paragraph with the links above and a line saying the DPA comes with the Commercial Terms is enough, and it’s what you’ll show a client who asks.
I give every team two rules: client secrets go only in a company seat, and nobody rates a chat that holds client material. Paste this to your team and fill in the brackets:
Client work goes only in our company Claude account, [Team / Enterprise], never in a personal one. Anthropic’s software checks chats for Usage Policy problems, and a flagged chat can be kept for years, so ask [name] before pasting [security scans, client medical records, anything a contract calls confidential]. [We’ve turned the thumbs up and down off. / Don’t click thumbs up or down on a chat that names a client: it sends the whole chat to Anthropic, where it can be kept for up to 5 years.] Incognito hides a chat from your history, not from Anthropic, and [name], our Primary Owner, can still export it. Deleting a chat takes it off Anthropic’s servers within 30 days, not on the spot. Questions go to [name].
What I couldn’t confirm
- Which team at Anthropic may read a Team or Enterprise chat, and when. The Commercial Terms limit it to employees with a need to know, but no page names the team or what triggers a look. If a client asks, put the question to Anthropic in writing and keep the answer.
- What the classifier flags. Anthropic publishes the rules, not the triggers. Keep security testing to systems you’re authorized on.
- How a legal hold works on Team. Nothing says. Assume anything typed can outlive its deletion.
What should I do next?
This week, turn off Rate chats, check every personal seat that touches client work, and send the note. If you’re choosing between Team and Enterprise because of retention, is Claude safe for business says when Team stops being enough.
If a client’s contract asks who at Anthropic can read their data, send Anthropic that question in writing before the work starts, and keep the reply with the vendor page from step 5. Setting Claude up for firms like yours is my job, so weigh my view on plans with that in mind: my setup turns the training setting off with you and gives your team those two rules, or bring the contract to a 30-minute call.
Sources for the facts on this page, checked November 7, 2026: How long do you store my organization's data?, How long do you store my data?, Is my data used for model training? (commercial), Is my data used for model training? (consumer), Configure custom data retention controls for Enterprise plans, Claude pricing, Commercial Terms of Service, HIPAA-ready Enterprise plans, Anthropic Trust Center.


