GMAsia
    AI 新聞·2026年8月26日·來源: Testing Catalog

    Claude unifies memory across Chat and Cowork

    內容只提供英文版本

    Anthropic is rolling out a unified memory system for Claude, integrating context across its chat interface and Cowork cloud task runner. This change, effective today, ensures that details from conversations, such as project priorities or team terminology, are automatically available when users transition to Cowork tasks. The new memory feature is enabled by default for Free, Pro, and Max web, desktop, and mobile users, while Team and Enterprise accounts require administrator activation. Users can manage stored information, organized by topic, within their settings, allowing for editing or deletion of individual memory files. Sensitive topics like health or political views are excluded by default but can be opted into with user consent, though Claude will still refuse to store highly sensitive personal identification numbers or criminal history.

    Nexa 摘要

    Anthropic's integration of a single memory across Claude Chat and Cowork marks a practical step in enterprise AI utility. This move addresses a common friction point for users who frequently switch between conversational AI and task execution, particularly in project management or content creation roles common across Asian tech firms. By automatically carrying context like project priorities or team terminology, Claude aims to reduce repetitive inputs and streamline workflows, which could improve adoption rates in corporate environments. The ability to edit and delete stored memory files, coupled with explicit opt-in for sensitive topics, provides a necessary layer of user control. This is crucial for Asian markets where data privacy regulations and cultural sensitivities around information sharing are significant. The system's refusal to store highly sensitive data like criminal history or identification numbers reflects a cautious approach to data governance, which will be watched closely by regional regulators and enterprise clients. The real test will be how effectively this memory system handles complex, multi-stage projects without introducing new layers of data management overhead for users.

    #anthropic
    原文報道: Testing Catalog我們不轉載全文, 閱讀原文 →

    相關文章

    6 則