Anthropic’s AI Will Now Share Your Memories—Here’s How To Opt Out

Anthropic has announced that user memories will now be shared automatically between its AI products Claude and Cowork by default, unless users choose to opt out. This shift in privacy settings affects anyone using both tools with memory enabled, raising questions about data control and privacy implications.

Previously, Anthropic’s memory feature required users to explicitly enable sharing of information between Claude, its conversational AI, and Cowork, its agentic work assistant. Now, the company states that memories such as preferences, ongoing projects, or personal details will automatically be accessible across both products unless users manually disable this feature in Claude’s settings.

The change is framed by Anthropic as a convenience feature, intended to reduce repetition and improve assistance by allowing seamless transfer of user context. For more details, see this coverage. Users can find the toggle to disable shared memory in Claude’s settings, but details about how the shared data is managed, whether it can be selectively pruned, or if existing memories migrate automatically, remain unspecified.

Implications of Default Memory Sharing

This change significantly impacts user privacy and data control. With default sharing enabled, most users will not need to take action, making cross-product memory sharing the norm. For privacy-conscious users, this raises concerns about personal information unintentionally flowing into work-related tasks or being retained without explicit consent. It also reflects broader industry trends where AI tools increasingly retain and transfer user data, prompting ongoing debates about data security and user autonomy.

Industry Trends and Previous Privacy Practices

Anthropic’s move aligns with industry-wide developments where companies like OpenAI and Google have expanded persistent memory features, sparking questions about how stored data is used, retained, and secured. Historically, many AI providers required explicit opt-in for memory sharing, but recent shifts toward default-enabled features suggest a trend toward more integrated, seamless AI experiences. The move also coincides with broader discussions about AI’s role in automating tasks and the associated privacy considerations.

“Users can opt out of memory sharing between Claude and Cowork in settings.”

— Anthropic spokesperson

Unanswered Questions About Data Management

Several key details remain unclear, including whether shared memories can be selectively pruned or if existing memories migrate automatically into Cowork. It is also unknown whether shared data is used for model training, and how deletions in one product affect the other. The timing of rollout and whether administrators can control these settings on shared accounts are also still being clarified by Anthropic.

Next Steps for Users and Industry Response

Users should review their Claude settings this week to disable memory sharing if desired. Anthropic is expected to release further details on selective sharing and data retention policies. Industry observers will watch for how competitors respond, especially regarding default privacy settings and user control, as AI memory features become a key battleground for trust and usability.

Key Questions

What exactly is being shared between Claude and Cowork?

Memories such as user preferences, ongoing projects, and personal details stored by Claude will be accessible to Cowork to improve task execution, unless the user disables sharing in settings.

How can I prevent my memories from being shared?

Go to Claude’s settings and toggle off the memory sharing option to keep your data separate across the two products.

Are existing memories automatically migrated to Cowork?

It is not yet clear whether current memories are automatically transferred or if they need to be manually managed. Anthropic has not provided specific details on this process.

Will shared memories be used for model training?

The company has not addressed whether shared data will be used to improve models or retained for training purposes, leaving this question open for future clarification.

Can I selectively share certain types of information?

At this time, it remains uncertain whether users can choose which memories to share or keep private on a category basis, or if the toggle is all-or-nothing.

Source: ThorstenMeyerAI.com