Skip to content
, ,

Is Claude Safe and Private? Data, Training and Settings (Nov 2026)

13–20 minutes
Surprised woman in glasses and an orange hoodie in a podcast studio holding a card with the Claude logo, hook text: 5 years or 30 days?

Is Claude safe and private to use in November 2026? Mostly yes, if you check a few settings. The biggest one is the model training toggle, because it decides whether your chats are kept for 30 days or up to 5 years.

Claude can also read your files, browse in Chrome and connect to apps now. That power brings new risks that simple chatbots never had.

This guide explains what Anthropic collects, how to opt out, and a step by step safety checklist you can finish in ten minutes.

TL;DR

  • Training: On Free, Pro, Max and Ultra, your chats are used for training only if the model improvement toggle is on. Turn it off in Settings > Privacy.
  • Retention: With training on, data can be kept (de-identified) for up to 5 years. With it off, deleted chats leave Anthropic’s back end within 30 days.
  • Incognito: The ghost icon starts a chat that skips history, memory and training. It is still kept for 30 days for safety.
  • Business: Team, Enterprise and API data is not used for training under the Commercial Terms. Zero data retention is available for eligible API use.
  • Age: Claude is for adults 18+. Accounts flagged as possibly under 18 must verify age through Yoti.
  • Agents: Claude in Chrome, Cowork and connectors can be tricked by prompt injection. Give them the smallest access you can.

Is Claude Safe? The Short Answer

Quick answer: Claude is a reasonably safe and private AI assistant if you turn off model training, avoid pasting secrets, and keep agent features on a short leash. Anthropic says it does not sell personal data, and business plans are not used for training.

“Safe” means different things to different people. So it helps to split the question into four parts.

  • Privacy: who sees your chats, how long they are stored, and whether they train future models.
  • Security: how well Anthropic protects your account and data from attackers.
  • Content safety: whether Claude gives harmful answers, and who is allowed to use it.
  • Agent safety: what can go wrong when Claude acts for you in a browser, on your files or inside connected apps.

Each part has its own settings and risks. The sections below go through them one by one.

What Data Anthropic Collects

Anthropic’s Privacy Policy (updated September 10, 2026) lists the personal data it collects. Here is the plain English version.

Data type Examples When it is collected
Account details Name, email, phone number When you sign up
Inputs and outputs Prompts, files, coding and agent sessions, Claude’s replies and actions Every time you use Claude
Payment info Card and billing details When you buy a paid plan
Device and technical data Device type, OS, browser, IP address, cookies, error logs Automatically
Location Rough location from IP; precise location only with consent Automatically or by consent
Usage data Access times, pages viewed, links clicked Automatically
Feedback Thumbs up or down, which can include the whole related chat When you rate a reply
Connector data Data Claude reads from or sends to linked apps When you enable a connector
Verification data Age check result or ID images Only if you verify

The policy says Anthropic does not “sell” personal data as defined by applicable laws. Depending on where you live, you can ask to access, correct, delete or port your data by emailing privacy@anthropic.com.

Note: Feedback is the sneaky one. A single thumbs up or down can attach the whole conversation, and that feedback data is kept for 5 years.

Does Claude Train on Your Chats? How to Opt Out

Since Anthropic’s consumer terms update (announced August 28, 2025), Free, Pro and Max users choose whether their chats help train future models. Every user had to pick a setting by October 8, 2025 to keep using Claude.

The choice applies to consumer accounts, including Claude Code used from those accounts. It does not apply to Team, Enterprise, Government, Education or the API.

How to turn off model training

  1. Open Settings. On the web or desktop app, click your name in the lower-left corner and choose Settings.
  2. Go to Privacy. Select Privacy, or open claude.ai/settings/data-privacy-controls directly.
  3. Find the model improvement toggle. Anthropic’s privacy center labels it “Help improve our AI models” (older screens said “Help improve Claude”).
  4. Switch it off. New chats and coding sessions will no longer be used for future training.
  5. On mobile, use the same path. In the app, tap your name, then Settings, then Privacy, and check the same toggle.

What opting out does and does not do

  • It stops future training: new and resumed chats are not used, and stored chats are left out of future runs.
  • It cannot undo the past: data already in a training run that has started, or in a model already trained, is not removed.
  • Safety flags still apply: chats flagged by safety classifiers may still be used to improve trust and safety systems.
  • Deleted chats are excluded: chats you delete are not used for future training, even with the toggle on.

How Long Claude Keeps Your Data

Retention depends on your settings and on what happens to the chat. These numbers come from Anthropic’s privacy center for consumer plans.

Situation How long it is kept
Model training on Up to 5 years, in de-identified form, in training pipelines
Model training off, chat deleted Removed from history at once, from back-end storage within 30 days
Incognito chat 30 days by default (longer only under an Enterprise custom policy)
Feedback (thumbs, bug reports) 5 years
Chat flagged for a Usage Policy violation Inputs and outputs up to 2 years; safety scores up to 7 years
Legal requirement or dispute As long as the law or case requires

Chats you never delete stay in your history until you remove them. So deleting old chats you no longer need is a simple privacy win.

Incognito Chats Explained

Incognito chats are available on every plan: Free, Pro, Max, Ultra, Team and Enterprise. They are the closest thing to a private mode in Claude.

How to start one

  1. Open a new chat outside a project. Incognito is not available inside projects.
  2. Click the ghost icon. It sits in the upper-right corner.
  3. Check the black border. An “Incognito chat” label appears in the upper left.
  4. Close it with the x. Once closed, it cannot be reopened.

What incognito does

  • Not in history: the chat does not appear in your chat list.
  • No memory: Claude does not use existing memory and does not save new memories from it.
  • No training: it is never used to improve models, even if the training toggle is on.
  • Still stored briefly: it is kept for 30 days for safety reasons.
  • Fewer tools: on the new Claude experience, incognito opens in the older interface, so Claude cannot create files or run code there.

Warning: On Team and Enterprise plans, incognito chats are included in organization data exports for Owners, and in the Compliance API on Enterprise. Incognito hides a chat from your history, not from your employer.

How to Export and Delete Chats or Your Account

Export your data first

  1. Use the web or desktop app. Exports cannot be started from the iOS or Android apps.
  2. Go to Settings > Privacy. Click “Export data”.
  3. Check your email. A download link arrives at your account email. It expires after 24 hours and may only work once.

Delete single chats

Open the chat list, use the menu next to a chat and choose delete. It leaves your history at once and Anthropic’s back end within 30 days.

Delete your whole account

  1. Cancel any paid plan. For Pro, Max or Ultra, cancel in Billing and wait for the current period to end.
  2. On the web: click your name, go to Settings > Account, then click “Delete account” and follow the prompts.
  3. On iOS: Settings, then Profile under Account, then “Delete account”, then “Delete”.
  4. On Android: open the menu, tap your initials, then Profile, then “Delete Account” under Account Actions, then “I Understand”.
  5. Team or Enterprise members: switch to the organization and use Settings > Account > “Close your personal account”.

Deletion is permanent. You lose chats, projects and any unused credits, so export first.

Team, Enterprise and API Data Terms

Business use runs under Anthropic’s Commercial Terms, and the rules are stricter. The key line says: “Anthropic may not train models on Customer Content from Services.”

Plan Used for training? Retention basics
Free, Pro, Max, Ultra Only if the toggle is on 30 days after deletion, or up to 5 years with training on
Team No (Commercial Terms) Chats kept so you can reuse them; owners control exports
Enterprise No (Commercial Terms) Custom retention controls, Compliance API, audit tools
API No (Commercial Terms) Inputs and outputs deleted on the back end within 30 days by default
API with ZDR No Prompts and replies not stored at rest after the response

What zero data retention (ZDR) covers

ZDR is set per organization and you request it through Anthropic sales. It covers the Messages and Token Counting APIs and Claude Code with commercial API keys.

It does not cover consumer plans, the Console playground, or Team and Enterprise chat interfaces. Some API features, such as the Files API, Batch processing, code execution and the MCP connector, keep data under their own rules.

Note: Anthropic’s docs say Covered Models (Fable 5 to 5.2 and Mythos 5 to 5.2) require 30-day retention even for ZDR organizations, unless Anthropic expressly agrees otherwise. Check this before sending regulated data to those models.

Admins who want tighter control, such as IP allowlists and SSO, should read my guide to Claude Enterprise security controls.

Security Certifications

Anthropic lists these certifications for its commercial products, like Claude for Work and the API.

  • SOC 2 Type I and Type II: independent audits of security controls.
  • ISO 27001:2022: information security management.
  • ISO/IEC 42001:2023: management systems for AI.
  • HIPAA-ready configuration: a Business Associate Agreement (BAA) is available for eligible customers.

You can request the audit reports at trust.anthropic.com. These certifications describe Anthropic’s own controls. They do not make it safe to paste passwords or patient records into a personal Pro account.

Is Claude Safe for Kids? Age Rules

Claude’s consumer app is for adults 18 and over. You confirm your age when you sign up, and Anthropic runs systems that look for signs of under-18 use.

  • If an account is flagged: it is disabled until the user passes an age check through Yoti, a third-party service.
  • Check options: a selfie age estimate, an ID document, or a verified “over 18” attribute from the Yoti Digital ID app.
  • What Anthropic sees: only a pass or fail. Yoti deletes the selfie or ID images after the check.
  • Mobile app stores: in some US states, if your App Store or Play Store account says you are under 18, you may not be able to sign in.

So Claude is not a kids’ app. Parents of teens should treat it like any adult service and use a family account setting on the phone to control installs.

Connectors, Claude in Chrome and Cowork Risks

The newest risks come from Claude acting for you. When Claude can read untrusted content and also take actions, a hidden instruction in a web page or email can hijack it. This is called prompt injection.

Claude in Chrome

  • Screenshots: Claude captures the tabs it works in, so anything visible can enter the chat.
  • Permission modes: “Manually approve” checks every action. “Automatically approve” runs a safety review and pauses when needed. “Skip all approvals” checks nothing.
  • Site permissions: choose “Allow this action” for one-off approval. “Always allow actions on this site” still asks before downloads or sensitive entries.
  • Blocked by default: adult and pirated content sites. Financial sites need your permission.

Anthropic recommends a separate browser profile with no banking, health or government logins. My Claude in Chrome guide covers setup in detail.

Cowork and connectors

  • File access: Cowork can read, write and delete files in folders you connect. Use one dedicated working folder and keep backups.
  • Delete protection: Cowork must ask for explicit permission before permanently deleting files, in any approval mode.
  • Connectors and MCP servers: each one adds a new attack path. Local MCP servers run with the same permissions as any program on your computer.
  • Network rules gap: egress permissions do not apply to web search, web fetch, MCPs or Claude in Chrome.
  • Your responsibility: Anthropic says you stay responsible for actions Claude takes for you.

New to Cowork? Start with what Claude Cowork is and how it works, then read my MCP security best practices.

Tip: Stop a task at once if Claude starts talking about unrelated topics, visits sites you did not ask for, or asks for a password. Those are classic signs of prompt injection.

Phishing and Fake Claude Apps

Scammers love popular AI brands. In July 2026, BleepingComputer reported a Bing ad campaign that pushed a fake “ClaudeDesktop.exe” installer. It installed SectopRAT, malware that steals passwords, cookies and card data.

The lure even used a malicious Claude Artifact hosted on Claude’s real domain, which Anthropic later removed. So a claude.ai link alone is not proof that a download is safe.

  • Official web app: type claude.ai yourself instead of clicking ads.
  • Desktop app: download only from Anthropic’s own download page, never from search ads or file-sharing sites.
  • Mobile: install the Claude app published by Anthropic in the App Store or Google Play.
  • Emails: be wary of messages saying your account is suspended or asking for payment details. Log in directly instead of clicking links.

Claude Safety Checklist (Step by Step)

Work through this list once, then review it every few months. It takes about ten minutes.

  1. Decide on training. Open Settings > Privacy and set the model improvement toggle the way you want. Off is the private choice.
  2. Secure your login. Use a strong, unique password or a Google or Apple sign-in protected by two-factor authentication.
  3. Clean old chats. Delete conversations with personal details you no longer need.
  4. Use incognito for sensitive one-offs. Health or money questions you do not want in history or memory belong there.
  5. Never paste secrets. Keep passwords, API keys, full card numbers and ID numbers out of every chat.
  6. Review memory. Check what Claude remembers about you and remove anything you do not want stored.
  7. Audit connectors. Disconnect apps you no longer use. Some integrations keep access until you disable them.
  8. Limit browser and file access. Use a separate Chrome profile and one dedicated Cowork folder. Revoke old site approvals in the extension’s Permissions page.
  9. Think before you rate. A thumbs up or down can attach the whole chat for 5 years.
  10. Download from official sources only. Ignore sponsored “Claude” ads.

If you are picking a plan partly for privacy, remember the training rules are the same on Free, Pro, Max and Ultra. See Claude Free vs Pro vs Max vs Ultra for the other differences.

Common Problems and Fixes

I cannot find the training toggle

Go straight to claude.ai/settings/data-privacy-controls while signed in. If you are on a Team or Enterprise account, there is no toggle because business data is not used for training.

My data export link does not work

The link expires after 24 hours and may work only once. Sign in to the same account, then run the export again from Settings > Privacy on the web or desktop.

The “Delete account” button is missing or greyed out

  • Paid plan still active: cancel in Billing and wait for the period to end.
  • Organization account: use “Close your personal account” instead.
  • Support-only account: your settings will say to contact support. Name every account you want removed if one email has several.

My account was disabled for age

Check your email for the Yoti verification link and complete the check. If the mobile app blocks you, your app store age may be set under 18, so ask Apple or Google support or the family group organiser.

Claude in Chrome keeps asking for permission

That is the safe default. For a trusted site, choose “Always allow actions on this site”, but leave banking and email on per-action approval.

I pasted a password or API key into a chat

Change the password or revoke the key right away. Then delete the chat, which removes it from Anthropic’s back end within 30 days.

Frequently Asked Questions

Does Anthropic read my Claude chats?

Anthropic’s policies describe chats being used for safety reviews when flagged, for feedback you submit, and where the law requires. Treat any cloud AI chat as something a reviewer could see.

Is Claude more private than ChatGPT?

Both offer a consumer training opt-out and separate business terms. In practice, your own settings matter most, so check the training toggle and memory on whichever tool you use.

Does a paid Pro or Max plan stop training on my data?

No. Free, Pro, Max and Ultra all follow the same consumer rule, so training depends only on your privacy toggle.

Are incognito chats deleted right away?

No. They skip your history, memory and training, but Anthropic keeps them for 30 days for safety.

Is the Claude API safe for customer data?

API data is not used for training, and inputs and outputs are deleted within 30 days by default. Businesses with stricter needs can request zero data retention or a BAA.

Can a 15-year-old use Claude?

No. The consumer app requires users to be 18 or older, and accounts flagged as possibly under 18 are disabled until age is verified.

Is Claude Cowork safe on my work computer?

It can be, with care. Connect only one dedicated folder, keep backups, use manual approval for sensitive tasks, and follow your company’s AI policy.

Sources

Free weekly email

Get new AI guides every week

One short weekly email with our newest local AI, ComfyUI and AI tool guides. No spam, no pop-ups, just the good stuff.

Free. Unsubscribe anytime in one click. We only use your email to send this newsletter. See our privacy policy.

Surprised woman in glasses and an orange hoodie in a podcast studio holding a card with the Claude logo, hook text: 5 years or 30 days?