Anthropic announces today that all paid Claude users will now have memory.
In September, the AI company announced that Claude memory was available to some paid users — those on Team and Enterprise plans. Now, Claude will be able to “remember” for individual Max and Pro users.
Max subscribers can enable Memory in Settings today, while Pro users will see it roll out in the coming days. When turning on the feature, you give Claude a starting point and it starts building context from there. You can also import memory from ChatGPT or Gemini by copy and pasting — and you can export memory out of Claude, as well.
Memory is optional and can be toggled on or off. Users can also delete specific memories or use Claude incognito (which the company also introduced recently).
In a press release shared with Mashable, Anthropic claims that it conducted safety tests with memory, such as whether Claude would recall conversations with harmful language or become overly accommodating with potentially harmful user requests (an issue that has been observed in other LLMs like ChatGPT, which has been criticized for feeding into some users’ delusional thinking). Anthropic states that it made adjustments to how memory functions, based on these tests.
Mashable Light Speed
When Anthropic released its latest lightweight model, Claude Haiku 4.5, it claimed it was its safest yet, and the company reiterated that its latest LLMs are safer than their predecessors.
In terms of memory specifically, the release states that Claude provides “complete transparency,” in that users can see the “actual synthesis, not vague summaries” of what the model stores.
Anthropic has released several updates to Claude in the past few months, including a new code model, the compact Haiku 4.5 model, a Chrome extension, and the ability to make spreadsheets and decks. Memory is powered by the Claude 4 model family, the company stated.
“We’re building toward Claude understanding your complete work context and adapting automatically,” Anthropic chief product officer Mike Krieger said in the press release. “Memory starts with project continuity, but it’s really about creating sustained thinking partnerships that evolve over weeks and months.”
Disclosure: Ziff Davis, Mashable’s parent company, in April filed a lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.
Topics
Artificial Intelligence
