Anthropic unifies Claude and Cowork memories, letting AI assistants share what they know about you

Anthropic’s frontier AI models have long been able to store user information — names of friends and colleagues, work preferences, interests — to provide more personalized responses. Now the company is extending that capability across its product line by linking the memory of its main Claude chatbot with Cowork, an emerging assistant designed to automate professional tasks beyond coding.

“Ask Cowork to draft an update for your manager, and it already knows who that is and how she likes updates written,” Anthropic said in its announcement. The company also described use cases such as brainstorming a conference agenda with Claude Chat and having Cowork automatically build budget and logistics documents using the same context, or defining a team’s preferred metrics format once in a chat and relying on Cowork to apply it quarterly.

The shared memory works bidirectionally: details from Claude Chat flow to Cowork and vice versa. Anthropic told The Register there is no option to keep the two assistants’ memories separate — the only way to isolate them is to use different accounts. Users can also pause memory entirely, use incognito chats for sensitive conversations, or manually delete individual memory items.

Memory generation itself has changed. Previously, Claude created memories by reviewing a conversation after it ended. Now memories are generated on the fly during chats. Anthropic emphasized that it does not store “sensitive” information — such as health, ethnicity, religious beliefs, politics, or gender identity — unless users explicitly opt in. If a user does enable sensitive memory, the data becomes available to Cowork as well.

For Team and Enterprise customers, administrators control whether the memory feature is enabled, meaning the integrated experience described by Anthropic will require IT approval. The company also confirmed that Claude Code, its coding-focused assistant, maintains a separate memory system for now, with no announcement about whether it will eventually join the shared memory pool.

The new shared memory went live on Tuesday. Cowork is not available on free accounts, so free-tier Claude users will create memories that only flow to Cowork if they later upgrade.

Why This Matters

User privacy under scrutiny: The default-on shared memory means personal information gathered by one tool is instantly accessible by another, reducing user control over data silos. The lack of a separation option forces users who want isolation to manage multiple accounts or disable memory entirely.
Productivity vs. surveillance tradeoff: For professionals, the convenience of AI that remembers context across tasks is significant, but it creates a single point of data exposure — any misuse or breach could leak a wide range of personal and work-related details.
Industry precedent: This move signals a broader trend among AI companies to unify memory across product families, raising questions about how sensitive data will be handled as models become more deeply embedded in daily workflows.

Background

Anthropic launched Claude in 2023 with limited personalization. Memory capabilities were introduced gradually, initially as a post-conversation review that stored explicit details. The company positioned memory as a tool to improve continuity, much like competitors OpenAI (ChatGPT) and Google (Gemini) have done with their own memory features. Cowork was introduced earlier in 2026 as a general-purpose work assistant distinct from Claude Code, which focuses on software development. The decision to share memory between the two reflects Anthropic’s strategy to create a seamless ecosystem, but it also echoes criticisms leveled at other assistants that aggregate user data without granular consent controls.

Key Perspectives

[Anthropic]: The company argues that shared memory dramatically improves workflow automation. “Cowork already knows who your manager is and how she likes updates written,” a spokesperson said, highlighting the reduction in repetitive context-setting. Anthropic maintains privacy safeguards by default-blocking sensitive categories and offering incognito and manual deletion options. [Privacy Advocates and Skeptics]: Critics note that default-on memory — especially with real-time generation — increases the risk of unintended data retention. The inability to separate Claude Chat and Cowork memories without separate accounts is seen as a flaw; users may accidentally share details meant for one domain (e.g., personal chat) with a work tool. They also point out that opt-in for sensitive topics is not the same as opt-in for all memory, and that Anthropic’s fly-on-the-wall monitoring must be extremely accurate to avoid capturing sensitive inferences from non-sensitive input. [Enterprise Customers]: For business clients, admin controls offer a safety valve — IT can disable memory entirely. However, the promise of AI that remembers employee preferences may be offset by compliance concerns, especially in regulated industries where data segregation is required. Some enterprises may prefer to keep memory off until Anthropic proves its real-time filtering can reliably exclude protected categories.

What to Watch

User backlash and opt-out rates: How many users disable memory or switch to incognito mode will signal whether the default-on approach is accepted or resisted.
Regulatory attention: Privacy regulators in Europe (GDPR) and California (CCPA) may scrutinize the real-time data collection and default sharing of memory across products.
Claude Code’s inclusion: Anthropic said it has nothing to share about joining the hive mind, but any future unification would expand the shared memory surface further, especially among developers.
Competitor responses: OpenAI and Google may follow suit with their own cross-product memory sharing, intensifying the debate around AI data practices.

ZOTPAPER