Superpower Daily
Claude Connects Chat Memory to Cowork, While Keeping Sensitive...OpenAI’s Jalapeño Claims 1.5–1.9x More AI Work Per Watt, Faces...Stanford Finds a 19% Employment Gap for Young Workers in AI-Exp...Nvidia’s NeMo Switchyard Routes Agent Calls Across Models, Not...Kimi.ai Uses TiDB for One-Second Agent Databases and Persistent...Two Planned English Data Centres Seek Gas Power, With a 4.577m-...Claude Connects Chat Memory to Cowork, While Keeping Sensitive...OpenAI’s Jalapeño Claims 1.5–1.9x More AI Work Per Watt, Faces...Stanford Finds a 19% Employment Gap for Young Workers in AI-Exp...Nvidia’s NeMo Switchyard Routes Agent Calls Across Models, Not...Kimi.ai Uses TiDB for One-Second Agent Databases and Persistent...Two Planned English Data Centres Seek Gas Power, With a 4.577m-...
Productspublished

Claude Connects Chat Memory to Cowork, While Keeping Sensitive Memory Opt-In

Claude’s new handoff promises less repeated context-setting across its products. The trade-off is a more active memory layer that users may need to inspect, edit, pause, or clear as projects change.

By 3 min read
Claude Connects Chat Memory to Cowork, While Keeping Sensitive Memory Opt-In
Sign in to save and react

Listen to this story

The audio brief

About 1:31
0:001:31
Read transcript
Claude can now carry what it learned in a conversation into Cowork, so turning a discussion into task work no longer necessarily means briefing it all over again. Anthropic says the shared context can include preferences, project details, and practical facts—like a conference’s city, speakers, or expected headcount. It can also carry writing-style preferences or metric definitions into later work. The bigger change is that memory updates while a conversation is still happening. A revised deadline, for example, can become available in a future Claude chat or Cowork task without the user separately asking Claude to remember it. That makes corrections more powerful, but it also raises the cost of stale or inaccurate context, because one saved detail can shape both products. Memories are organized into topic files. Users can inspect, edit, or delete individual entries, and they can pause or reset memory more broadly. Memory is enabled by default for free, Pro, and Max users across the web, desktop, and mobile apps, including iOS and Android. Sensitive-topic memory is different. Health, race, ethnicity, religion, politics, and gender identity are excluded unless the user explicitly turns that option on, and Claude shows a notice when it saves related information. Even then, government IDs, financial account numbers, criminal history, and immigration status remain excluded. The key thing to watch is whether these controls are enough as one evolving memory layer starts influencing both conversation and execution.

Story brief

3 key points

Anthropic is broadening Claude’s memory from a chat feature into shared context for Cowork, with updates captured during conversations rather than only afterward. Saved topic files can be edited, deleted, paused, or reset, and corrections apply across both products. Memory is on by default for free, Pro, and Max users on web, desktop, and mobile. Sensitive-topic retention remains opt-in, while government IDs,...

  1. 01

    Memory is enabled by default across Claude’s free, Pro, and Max plans on web, desktop, and mobile.

  2. 02

    Sensitive-topic memory requires explicit activation and a notice appears when Claude saves related information.

  3. 03

    Users can manage memories by topic file, including editing individual entries or pausing and resetting memory globally.

Anthropic is making Claude’s stored context travel from chat into Cowork, removing a repeat briefing when a conversation turns into task work. But Claude now also updates memories during chats, making its topic-level controls and sensitive-topic defaults more consequential.

A handoff that keeps the prior context

Cowork can now use stored memory from earlier Claude conversations, including preferences and information from prior projects. A team that has explained its metric definitions in chat, for example, can carry that context into a Cowork assignment rather than supply the same briefing again.

The shared context is intended for the details that often get lost between planning and execution. If a user discusses a conference in chat, Cowork can carry over details such as its city, speakers, and headcount when working on logistics. It can also retain writing-style preferences and previous project information for later work.

The memory layer also updates while a conversation is underway, rather than waiting for a chat to end. Users do not have to explicitly ask Claude to remember a detail; a changed project deadline can carry into a later conversation without a separate instruction.

Memory becomes a settings surface

Saved memories appear as individual files organized by topic. Users can read, edit, or delete those files, and Claude’s memory settings also let them pause or reset memory altogether. That gives users separate ways to correct a topic, temporarily stop retention, or clear saved context.

The controls matter more when remembered context can reach another product and change during a live conversation. Editing a topic changes the information available to future chats and Cowork tasks; pausing and resetting provide broader stops when a user does not want that stored context to persist.

Sensitive topics are optional, with fixed exclusions

Claude does not save health, race, ethnicity, religious beliefs, politics, or gender identity by default. Users can enable memory for those sensitive topics, and Claude displays a notice when it saves information involving them. Sensitive-topic memory remains off unless the user turns it on.

The opt-in can make information such as a dietary restriction available for relevant future requests, including meal planning. Yet the setting does not create an open-ended store for every sensitive record: Anthropic keeps several categories outside memory even when a user has enabled the option.

Records Claude will not save

  • Government ID numbers and financial account numbers remain excluded, even when sensitive-topic memory is enabled.
  • Criminal history and immigration status are also excluded under that setting.
  • Claude notifies users when it cannot save one of those categories.

The default reaches every plan

Memory is enabled by default on free, Pro, and Max plans across web, desktop, and mobile; sensitive-topic memory requires a manual choice. Anthropic is also updating its iOS and Android apps with the latest memory features, extending the same change to Claude’s mobile users.

Sources

  1. cnet.comAnthropic Unifies Memory Across Claude and Cowork - CNET
  2. digitaltrends.comClaude now remembers what you told it, wherever you use it

Related coverage

Continue reading