Anthropic Updates Claude Memory Controls and Chat Search
Anthropic updated Claude's memory experience on August 25, adding topic-level review, editing and deletion controls while excluding sensitive subjects from memory by default, SiliconANGLE reports. The update carries remembered context across Claude chats and Cowork tasks. Anthropic's Help Center also documents chat search, a RAG-based feature available to paid-plan users that retrieves relevant prior conversations through visible tool calls.
Anthropic updated Claude's memory experience on August 25, adding controls that let users review remembered information by topic, edit it or delete it, according to SiliconANGLE. The publication reports that Claude does not store sensitive subjects by default, including health concerns, race, ethnicity, gender identity, religion and political beliefs; users can enable storage of those topics with a toggle.
The update extends Claude's remembered context across ordinary chats and Cowork, Anthropic's desktop agent for multistep work, SiliconANGLE reports. The publication describes examples in which context about reporting preferences, metric definitions and visualization conventions can be reused when Cowork produces documents or slide decks.
Search and memory use separate mechanisms
Anthropic's Help Center describes a related chat-search capability that lets users ask Claude to find and reference relevant information from prior conversations. The feature is available on paid Pro, Max, Team and Enterprise plans across the web, Claude Desktop and Claude Mobile, according to the documentation.
The Help Center states that chat search uses retrieval-augmented generation, or RAG, and appears in a conversation as a tool call. Searches can cover chats outside projects or conversations within an individual project, but project searches remain bounded to that project. Once the capability reaches an account, Anthropic documents it as enabled by default, with a settings toggle to disable search and reference of prior chats.
Migration and privacy controls
Anthropic has introduced an improved memory experience and migrated users from a legacy version, the Help Center states. Users who believe information was lost during migration can export legacy memory until September 9, 2026, then paste relevant material back into Claude.
For teams evaluating persistent assistant context, the distinction between RAG-based retrieval and saved memory matters operationally. Comparable systems require separate governance decisions for what can be retrieved from historical conversations, what is retained as durable context, and which controls users and administrators can audit. Topic-level editing and explicit sensitive-topic defaults provide more granular controls than a single all-or-nothing history setting, though organizations handling regulated data still need to assess their own retention, access-control and data-governance requirements.
Key Points
- 1Anthropic added topic-level memory review and deletion controls, giving Claude users more direct visibility into retained conversational context.
- 2Claude chat search uses RAG and visible tool calls, separating retrieval of prior discussions from durable memory behavior.
- 3Comparable persistent-context systems require distinct governance for retrieval, retention, sensitive data defaults and user-controlled deletion.
Scoring Rationale
The update affects how Claude users manage persistent context across chat and agent workflows, a practical concern for enterprise AI deployments. Its RAG-based chat search and memory controls are relevant to teams designing privacy-aware assistant experiences, although this is an incremental product update rather than a new model release.
Sources
Primary source and supporting public references used for this report.
Practice with real FinTech & Trading data
90 SQL & Python problems · 15 industry datasets
250 free problems · No credit card
See all FinTech & Trading problems
