Anthropic has announced the expansion of its memory feature to all paid users of its AI model, Claude. Previously, memory was only available to those on the Team and Enterprise plans. Now, individual users subscribed to the Max and Pro tiers can enable this feature, allowing Claude to retain context from previous interactions.
Max subscribers can activate the memory function in their settings today, while Pro users will see the rollout in the coming days. When users turn on memory, they provide Claude with a starting point that the AI uses to build context over time. Additionally, users have the option to import memory from other AI models, such as ChatGPT or Gemini, by copying and pasting, and they can also export their memory from Claude.
The memory feature is designed to be optional, allowing users to toggle it on or off as needed. Users can delete specific memories or opt to use Claude in incognito mode, a feature introduced recently. In a press release shared with Mashable, Anthropic emphasized that it conducted safety tests to ensure Claude would not recall harmful conversations or overly accommodate potentially dangerous user requests, an issue that has been noted with other large language models (LLMs).
Mike Krieger, chief product officer at Anthropic, highlighted the company’s commitment to safety, stating that adjustments were made to memory functionality based on the results of these tests. The company claims that its latest lightweight model, Claude Haiku 4.5, is the safest iteration yet. Anthropic maintains that the newer LLMs are designed to be safer than their predecessors.
With regard to memory, Anthropic asserts that there is “complete transparency,” allowing users to view the “actual synthesis” of information stored by Claude, rather than vague summaries. This level of transparency aims to bolster user trust in the AI’s capabilities.
In recent months, Anthropic has rolled out several updates to Claude, which include a new code model, the compact Haiku 4.5 model, a Chrome extension, and features that allow users to create spreadsheets and presentation decks. Memory functionality is powered by the Claude 4 model family, further enhancing the user experience.
Krieger articulated a vision for the future, stating, “We’re building toward Claude understanding your complete work context and adapting automatically.” He noted that while memory begins with project continuity, the ultimate goal is to foster sustained thinking partnerships that evolve over weeks and months.
In a related note, it is important to mention that Ziff Davis, the parent company of Mashable, filed a lawsuit against OpenAI in April 2023, alleging copyright infringement in the training and operation of its AI systems. This legal action underscores the ongoing complexities surrounding the development and use of AI technologies.
