‍AI chatbots are increasingly equipped with⁤ the ability to retain information from past conversations,a feature known as “memory.” This development, while enhancing user experience, is simultaneously sparking concerns about ⁢data privacy and the potential for chatbots to generate inaccurate or misleading ‍responses-sometimes described as “delusional.”

The Rise of AI Memory

Anthropic announced on September 11, 2024, that its Claude chatbot ⁤will now retain details from previous interactions, allowing for more contextually relevant and ⁣personalized conversations according to ‍a company blog post. Google has already launched similar cross-chat memories⁣ for ⁢its Gemini chatbot as reported by The Verge.

This move towards persistent memory⁢ marks a⁣ significant evolution in chatbot technology. Previously, each conversation was largely treated as a discrete‍ event. Now, chatbots can build a profile of‍ user⁤ preferences, past requests, and ongoing projects, leading to more ⁢efficient and tailored interactions.

Privacy Concerns and the Introduction of Controls

⁣ Alongside the rollout of memory ⁣features, companies are responding to user privacy concerns by introducing controls. anthropic is offering “incognito chats” for claude, ensuring conversations aren’t saved or referenced in⁣ future interactions. ⁢ Similarly, Google introduced “Private Chats” for Gemini in‍ August 2024. these features allow ⁣users⁢ to have confidential conversations without contributing to the ‍chatbot’s long-term memory.
⁣ ⁤

The “Delusion” Factor: A Potential Downside

⁢ The implementation of cross-chat memories isn’t without⁢ potential drawbacks. A report published in The ⁣New⁤ york times on August 8, ‍2025, ⁣ linked ChatGPT’s rollout of cross-chat memories to an increase in reports⁣ of “delusional” AI chats. This‍ refers to instances where ‍chatbots ⁢confidently present fabricated information or exhibit illogical reasoning, possibly stemming from misinterpreting or conflating information stored in their memory.

The risk of “hallucinations” – where⁤ AI generates factually incorrect information – is ⁢amplified when chatbots have access ⁣to a broader range of past interactions. ‍ The ability to draw connections across conversations can⁢ lead to the creation of narratives that are internally ‍consistent but detached from reality.
⁣

Comparing Memory Features: Claude, Gemini, and ChatGPT

While all three major chatbots are incorporating memory, their⁣ approaches differ. Here’s a brief comparison: