Claude and Your Data: Anthropic Confronts Privacy

Claude and Data Privacy: An Opaque Management

Many users entrust Claude with information they would not easily share publicly, such as medical inquiries, legal issues, or sensitive professional data. However, the way Anthropic actually manages these exchanges remains quite unclear. The levels of privacy vary, and the default settings are not always the most protective.

Depending on the type of subscription and the chosen settings, Anthropic processes data differently. Let’s examine what changes and what users can do to protect their information.

Conversation Retention: A Default Retention Policy

By default, claude.ai retains the entire conversation history. Anthropic specifies in its privacy policy that these exchanges are stored on its servers located in the United States. Internal teams can access them in specific contexts, such as:

• Security
• Service improvement
• Detection of problematic behaviors

While it is possible to manually delete a conversation from the history, this does not result in immediate deletion at the infrastructure level. Conversations remain on the servers for 30 days before being permanently deleted.

This rule is not set in stone. Data processing depends on how the service is used. For example, a developer using the API from Anthropic to create a third-party application or service is not subject to the same rules as a user on claude.ai. With the API, data is not used to train the models by default, and log retention has been reduced to 7 days since September 2025. In contrast, Free or Pro subscribers on claude.ai are subject to a more exposed regime.

Claude's Memory: Customization and Deletion

Claude also has a "Memory", distinct from the history, which retains information between sessions to personalize future interactions. If you have mentioned your profession, work habits, city, or response format preferences, Claude can keep these elements and reuse them in future conversations.

These memories are generated automatically or can be added manually by the user. They accumulate without apparent limits and without systematic notifications. To view them, simply go to your account settings, under the "Memory" section.

If you prefer that Claude does not retain anything between sessions, it is possible to completely disable this feature in the settings by toggling the "Memory" option to "Off." From there, each conversation starts from scratch.

⚠️ Two important points:

• Disabling does not delete already stored memories. This step must be done separately before disabling the feature.
• Projects, which allow Claude to have permanent context on a given topic, have their own retention logic and are not affected by the global memory deactivation.

Data Usage for Model Training

This is the most significant change to be aware of. During its first two years, Anthropic distinguished itself from its competitors by not using conversations from its general users to train its models. This promise ended on September 28, 2025. Since that date, for Free, Pro, and Max accounts, conversations are used by default to improve future versions of Claude.

To stop this behavior, you need to go to "Settings" > "Privacy" and disable the option related to model improvement. Anthropic then commits to not using your future conversations to train its models.

However, there is an exception: conversations flagged for security review may still be utilized. Conversations you manually delete are erased from the systems within 30 days. If the option remains active, Anthropic can retain your data for up to… five years.

Incognito Mode and Alternatives: Options to Explore

Since September 2025, Claude offers an incognito mode, accessible to all users, including those on the free plan. To activate it, simply click on the ghost icon when starting a new conversation. Once the window is closed, the conversation disappears from the history and is not integrated into the memory.

However, this mode has limitations. On one hand, it is not compatible with Projects. Moreover, contrary to what the term "incognito" might suggest, a 30-day retention still exists on the server side for security reasons.

To go further without creating an Anthropic account, DuckDuckGo AI Chat offers access to Claude through its anonymous chat interface. No conversations are stored by DuckDuckGo, no personal data is collected, and metadata is deleted within 30 days on the Anthropic side, with no use for training.

The Limits of Privacy Settings

None of these options turn Claude into a fully private service. Anthropic is an American company subject to U.S. law, which implies the possibility of a legal injunction for data access under certain circumstances. End-to-end encryption, as known with Signal, remains difficult to implement.