Claude
Published November 14, 2026· Updated November 14, 2026· By Juan Martinez
In This Article
Claude is the AI chat app of Anthropic, the San Francisco company that also makes the models behind it. It runs in a browser at claude.ai and in apps for Windows, macOS, iPhone and Android, with a Linux app in beta. Every plan, from Free to Ultra, needs an account, and a new account needs a mobile phone number.
Chats are stored on Anthropic’s servers, encrypted, in a form Anthropic can read. Whether they also train future models depends on one setting, and Anthropic’s own pricing page calls the arrangement opt-out. With training on, Anthropic keeps chats for up to five years; with it off, a deleted chat leaves its systems within 30 days.
Anthropic has promised to keep Claude free of ads, and its privacy pages give retention periods in plain numbers. The apps and the models are closed source, and the company’s security certifications cover its business products, not the personal plans.
Privacy Architecture
Claude does not work without an account (account required), and there is no guest chat. Every new account must also confirm a phone number that receives text messages. VoIP numbers, Google Voice and landlines are refused, the step cannot be skipped, and the number cannot be changed later. Anthropic says it uses the number only for verification, to stop spam and abuse. Accounts are open only to people who are physically in a country Anthropic supports.
Anthropic’s privacy policy lists what the apps send on their own: device and browser details, the IP address and the location worked out from it, device and advertising identifiers, and usage and log data. Anthropic runs the app and makes the model, so no second company receives the request, but nothing in its documents describes removing the IP address from it (IP anonymization).
Chats are encrypted in transit and on Anthropic’s servers. That protects them from a stolen disk, not from Anthropic. Its staff cannot open conversations by default, but designated members of the Trust & Safety team can when a chat needs review under the Usage Policy, and a chat sent as feedback is readable too. This is storage that the operator can decrypt.
The training setting decides most of the rest (provider training claim). Since the consumer terms changed in August 2025, Anthropic trains new models on chats and coding sessions from personal accounts when the switch “Help improve our AI models” under Settings, Privacy is on. New users choose at sign-up. Existing users were shown a pop-up, and TechCrunch reported that its training switch sat under a large Accept button and was already turned on. The pricing page lists model training as “Opt-out” on Free, Pro, Max and Ultra.
With training on, Anthropic may keep chats in a de-identified form for up to five years in its training pipelines. Turning the switch off removes old and new chats from future training, but chats already in a training run or in a finished model stay there. Two paths lead to training even with the switch off. Chats that Anthropic’s safety classifiers flag can be used to improve its safety models. A thumbs-up or thumbs-down rating sends the whole conversation to Anthropic as feedback, which it keeps for five years. Files and messages that connected apps such as Google Drive pass to Claude are not used for training unless they are pasted into the chat.
Incognito chats, opened with the ghost icon, are not saved to the history or to memory and are not used for training, even when the training switch is on. Anthropic still keeps them for 30 days. They work only outside projects and cannot be turned into a regular chat.
Some account data leaves Anthropic through the website. A May 2026 study by researchers at IMDEA Networks and UC3M in Spain tested how four AI chat services pass data to other companies. On Claude’s website, when a signed-in user accepts non-essential cookies, Anthropic’s servers forward the account email, plan, country and the page address with the conversation’s ID to eleven advertising and analytics services, among them the conversion tools of Facebook, LinkedIn, TikTok and Google. Rejecting those cookies stopped the forwarding. The support-chat tool Intercom received the open conversation’s address together with the user’s email whatever the cookie choice. The items the researchers list are identifiers and addresses, not the text of the chat. Anthropic’s privacy policy says it does not sell personal data, lets users opt out of targeted advertising for Anthropic’s own products and honours Global Privacy Control.
Trust & Transparency
Anthropic, PBC is a Delaware public benefit corporation, a legal form that lets its directors weigh a stated public benefit against shareholder returns. It is the data controller for users outside Europe; for the EEA, the UK and Switzerland the controller is Anthropic Ireland, Limited in Dublin. The consumer terms follow California law, with disputes heard in San Francisco courts. US legal process therefore applies to the chats Anthropic stores, and the US is part of the Fourteen Eyes intelligence-sharing group.
The privacy centre gives retention in numbers. A deleted chat leaves the history at once and the back-end storage within 30 days. Chats that the automated safety systems flag as breaking the Usage Policy are kept for up to two years, and their safety scores for up to seven. Feedback is kept for five years, and any chat can be kept longer when the law requires it or a dispute needs it. To delete an account on a paid plan, the subscription has to be cancelled and run out first.
Anthropic’s Trust Center lists a SOC 2 Type II report and ISO 27001 and ISO 42001 certificates, among others. The table that shows what they cover names the API, Claude Team, Claude Enterprise and the versions run by cloud partners; Free, Pro, Max and Ultra are not in it, and the reports themselves need an access request. For a personal account there is no outside audit to point to, and every privacy statement rests on Anthropic’s own documents. The apps and the models are closed source (open source).
No breach of personal Claude accounts or chats is on record. In February 2026 Anthropic announced that Claude will stay ad-free, with no sponsored links beside conversations and no answers shaped by advertisers. Usage and log data are collected with no switch to stop them, while the cookie-based analytics on the website can be refused (telemetry).
Features
Claude runs four Anthropic models, picked from a menu (multi-model switching). Free includes Claude Sonnet 5.5 and Claude Haiku 5.5. Pro adds Claude Opus 5.5 and Claude Fable 5.2, Anthropic’s top model, which Pro users pay for with usage credits; Max and Ultra include Fable for up to half of the weekly limit.
Every plan has web search, voice mode, artifacts, file creation with code execution, connectors to other apps, memory and incognito chats. File upload takes PDF, Word, spreadsheet, text, e-book and image files, up to 500 MB per file and 20 files per chat. PDFs can run to 1,000 pages, and Claude reads their charts and images only in files of 100 pages or fewer. Paid plans add Research, projects without the five-project cap, search across past chats, scheduled tasks, Claude Code, Claude in Chrome and add-ins for Microsoft 365.
Memory is on by default on personal plans. Claude saves facts from chats as separate topics and uses them in new conversations; the topics can be reviewed and deleted, and memory can be switched off for one chat from the “+” menu. Connectors let Claude read and act in services such as Google Drive and Slack, and each connected service handles what Claude sends it under its own privacy policy.
Usability
Claude works in a browser and in apps for macOS, Windows (including Windows on Arm), iPhone and Android. A Linux app is in beta for Ubuntu 22.04 and Debian 12 or later, without dictation or computer control. Chats, projects and memory sync across devices on one account.
Getting started takes more steps than opening a page and typing. A first message needs an account, a phone number that receives texts and a location in a supported country. Use is capped by a limit that resets every five hours and, on paid plans, by a weekly limit as well. The cap depends on the model, the length of the chat and the size of the files, so there is no fixed message count to plan around.
Price & Value
The Free plan covers everyday chat on Sonnet and Haiku with web search, file uploads, memory, connectors and up to five projects, under a limit that resets every five hours and every week. It has no Opus or Fable, no Research and no search across past chats.
Pro costs $20 a month, or $17 a month when $200 is paid for a year up front, and Anthropic says it gives at least five times the Free allowance per five-hour window. Max costs $100 a month for five times Pro’s usage. Ultra costs $200 a month for ten times Pro’s weekly usage, with no five-hour limit for verified accounts. Max and Ultra are billed monthly only and include $100 and $200 a month in credits for Anthropic’s developer platform. Prices exclude tax.
Paying does not change how data is handled: training stays opt-out and storage stays the same from Free to Ultra.
Pros and Cons
- Four Anthropic models, from the small Haiku 5.5 to the top Fable 5.2
- Web search, voice mode, artifacts, code execution, connectors and memory on every plan, including Free
- File uploads up to 500 MB per file and 20 files per chat
- Apps for Windows, macOS, iPhone and Android, plus a Linux app in beta
- Incognito chats are kept out of history, memory and training
- Retention periods are published in plain numbers: 30 days after deletion, five years with training on, two years for flagged chats
- Anthropic has committed to keeping Claude free of ads
- Memory can be reviewed, edited and switched off for a single chat
- Chats train future models unless the user switches training off, and the 2025 pop-up showed that switch already on
- With training on, chats can be kept for up to five years
- Flagged chats and chats rated with the thumbs buttons can be used even with training off
- An account and a phone number that receives texts are required, and the number cannot be changed
- Chats are stored in a form Anthropic can read, and Trust & Safety staff can open flagged ones
- The IP address and device identifiers are collected with the account
- With cookies accepted, the website forwards account email, plan and conversation IDs to eleven advertising and analytics services
- The support-chat tool receives the conversation address and email regardless of cookie choice
- Security certifications cover business plans and the API, not personal accounts
- Closed-source apps and models
- Usage limits reset every five hours with no fixed message count
- Heavy use costs $100 or $200 a month
Our Rating
Claude suits someone who wants strong Anthropic models with search, files, memory and connected apps on every major system, and who will switch training off and leave the thumbs buttons alone. It suits someone less well if they want to chat without an account or a phone number, or need chats that the company running the service cannot read. Anthropic states its retention periods clearly, but the protections it offers depend on settings the user has to change.
- Privacy Architecture1.5/10
Account and phone number required, IP collected, chats readable by Anthropic, and training on unless switched off.
- Trust & Transparency5.0/10
Retention is published in numbers and no breach is on record, but certifications skip personal plans and the website feeds ad platforms.
- Features8.5/10
Four models with search, large files, Research, memory, connectors, artifacts and code execution.
- Usability7.5/10
Apps for every major system, held back by phone verification and usage caps that reset every five hours.
- Price & Value6.5/10
Usable free plan and Pro at $20, but heavy use costs $100 or $200 a month.
Privacy Architecture 30% · Trust & Transparency 20% · Features 20% · Usability 20% · Price & Value 10%