"Anthropic collects extensive sensitive data including biometrics and government IDs, with AI model training by default and broad third-party sharing. Default model training and exceptions to opt-out undermine user control."
Take Action
5 Privacy Improvements Available
anthropic Privacy Concerns & Scorecard
Anthropic collects extensive sensitive data including biometric data (facial geometry, photos, videos), government-issued identity documents, precise device location, health app integration data, all chat inputs and outputs, uploaded files, and device identifiers. This represents 33+ distinct data types with multiple sensitive categories.
While opt-out mechanisms exist for model training and targeted advertising, exceptions undermine control. Flagged and user-reported conversations still train models despite opt-out. Policy contains vague language on data handling. Third-party integration data practices not controlled by Anthropic.
Data shared with affiliates, corporate partners, service providers, employer organizations (for enterprise accounts), third-party service integrations that receive Inputs/Outputs, government authorities and law enforcement, and may be transferred in corporate transactions. Data transferred globally outside EEA/UK to US and other countries.
Cookies and similar technologies used to recognize users, customize experience, and market additional Anthropic products. Device and advertising identifiers collected automatically. Data processing for direct marketing based on legitimate interests. Opt-out available but not enabled by default.
Data retention policy uses vague language (as long as necessary). Deleted conversations removed from history but retained in backend for 30 days. Data may be retained longer for legal, contractual, and regulatory compliance. No clear guarantee of deletion after purpose served.
Data retention training pipeline
- 1Navigate to your Claude account Settings.
- 2Select Privacy and turn off the 'Help Improve our AI models' toggle to stop new chats from entering training pipelines.
- 3Optionally, delete individual chats from your chat history to prevent those specific conversations from being used in future model training.
Conversation deletion
- 1Open the conversation you want to delete in Claude.
- 2Click the three-dot menu on the conversation.
- 3Select 'Delete conversation' to remove it from your history and queue it for deletion from back-end systems within 30 days.
Ai training opt out
- 1Select your name from the settings menu and click the drop-down button.
- 2Select Settings, or navigate directly to the Claude settings page.
- 3Select Privacy, or navigate directly to the data privacy controls page.
- 4Under 'Help Improve our AI models', click the toggle button to turn it off.
Automate your privacy
Connect your accounts to TrueTerms to automate these privacy actions and monitor policy changes.
Data Collection & Tracking
Personal Information Collected
Account creation, identity verification, age verification, and compliance with regulatory obligations.
Account management, customer support, direct marketing, and service communications.
Identity verification and age verification for account access and compliance.
Compliance with regional privacy laws, service personalization, and fraud prevention.
Processing payments, billing, and fraud prevention.
Service delivery, personalization, analytics, and technical support.
Service improvement, AI model training (by default), customer support, and safety review.
User recognition, personalization, analytics, targeted advertising, and marketing.
Product research, safety research, and service improvement.
Service personalization and feature enablement when health app integrations are enabled.
Enterprise account management, billing, and compliance for organization-wide deployments.
Cross-Platform Tracking
Anthropic uses cookies, device identifiers, and advertising identifiers to recognize users across sessions and potentially across websites where Claude is embedded or accessed. Device fingerprinting and cross-device tracking are not explicitly described, but the collection of device type, operating system, browser, IP address, and advertising identifiers enables cross-platform user recognition for personalization and marketing.
Tracking Methods:
First-party Cookies
Cookies placed by Anthropic to recognize logged-in users, maintain session state, and personalize experience within Anthropic services.
Third-party Cookies & Tracking Pixels
Third-party cookies and tracking pixels used for analytics, user recognition across sessions, and targeted advertising for Anthropic products.
Device Identifiers & Advertising IDs
Automatic collection of device advertising identifiers and unique device identifiers to recognize users across sessions and enable cross-device tracking for personalization and marketing.
IP Address Tracking
Collection and logging of IP addresses for compliance (regional privacy laws), fraud prevention, and analytics.
Account Login Tracking
Session tracking and login records to monitor user access patterns and authenticate users across devices.
AI & Data Training
anthropic uses your content to train AI
Below is what's used and how (if at all) you can object.
What Content Is Used
Chat Inputs and Outputs
All user inputs and Claude outputs are used for model training and improvement by default. Data is disassociated from user ID after training use, but personal data in inputs may appear in outputs and be retained for training.
Flagged Conversations
Conversations flagged for safety review are used for model training regardless of user opt-out status. This exception means safety-flagged content enters training data even if user has disabled general model training.
User-Reported Content
Content explicitly reported by users (e.g., for abuse, violation of terms) is used for model training regardless of opt-out status, to improve safety and compliance models.
Uploaded Files and Documents
Files uploaded to Claude are collected and may be used for model training unless user has opted out. Personal data in documents may be reproduced in model outputs.
Feedback and Ratings
User feedback, ratings, and thumbs-up/down signals provided during conversations are collected and used for model improvement and training.
How to Object
- 1
Users can opt out of having Inputs and Outputs used to train models through account settings in Privacy Settings. However, this opt-out does not apply to conversations flagged for safety review or explicitly reported by users.
Why We Analyzed anthropic
TrueTerms automatically audits privacy policies and data practices using advanced machine learning to keep you informed and protected. This scorecard is based on the latest available public terms of service and privacy policies as of 2026.