Anthropic updates Claude memory controls and unifies history across tools

Anthropic announced that its AI assistant, Claude, now shares context seamlessly between its regular chat interface and its task automation environment, Claude Cowork. According to the company's blog post, background knowledge gathered in chat immediately carries over to automated tasks and vice versa. Rather than writing a summary after a session ends, the system updates its saved topics in real time during ongoing interactions across web, desktop, and mobile devices.

Users maintain direct control over stored information through a settings panel where individual memory files can be reviewed, edited, or deleted. To protect privacy, Anthropic stated that subjects related to personal health, political beliefs, and demographic information are excluded from memory by default. People can manually toggle these sensitive categories on for customized assistance, such as tracking food allergies, though high risk identifiers like government identification numbers remain strictly off limits.

This unified memory model removes the friction of re-explaining project details or formatting preferences every time a person starts a new prompt. By giving individuals clear oversight over what the model remembers, software creators aim to make continuous AI context feel more useful and less intrusive.

However, centralizing detailed personal context across multiple tools creates new security responsibilities for both software makers and businesses. As digital assistants collect deeper organizational knowledge to handle complex chores, questions remain about how effectively these saved memories will be guarded against unauthorized extraction or malicious prompt injection attacks.