In a move to catch up with competitors like OpenAI and Anthropic, French company Mistral AI has announced a new “Memories” feature for its chatbot, Le Chat. This optional feature allows the bot to remember personal details from past conversations to deliver more personalized responses, raising questions about privacy and accuracy.
Article contents:
- Introduction: Artificial intelligence becomes more personal
- How does the “Memories” feature work?
- Privacy concerns: Mistral’s European approach
- The accuracy issue: Is “may remember” enough?
- The other new feature: “Connectors” to link services
- Security risks of connectors
- Conclusion: The delicate balance between personalization and security
Introduction: Artificial intelligence becomes more personal
Paris-based Mistral AI announced on Tuesday the introduction of a beta version of the “Memories” feature for its Le Chat chatbot, the company’s answer to ChatGPT. This feature, already available from competitors, allows the chatbot to be more useful by storing stated preferences and details from previous interactions to guide future responses.
How does the “Memories” feature work?
The idea is simple yet powerful. For example, the company suggests that if you tell Le Chat that you have a peanut allergy, “Le Chat may remember that to exclude peanuts from recipe suggestions.” This transforms the chatbot from a memoryless tool that answers each question separately into a personal assistant that learns about you over time.
Privacy concerns: Mistral’s European approach
This type of personalization raises the same potential privacy concerns as it does in search and advertising. Perhaps because Mistral operates out of Europe where strict data regulations exist (such as GDPR), the company published a detailed explanation in its privacy policy regarding how it uses this data and the options available for customers to control it. The company emphasizes that the “Memories” feature is an opt-in service.
The accuracy issue: Is “may remember” enough?
There is a subtle but important distinction in the company’s wording. Its documentation indicates that Le Chat “may remember” rather than “will remember.” The company’s post on the subject notes that Le Chat’s chance of accurately retrieving saved information is 86%. This is not a guaranteed percentage. Therefore, those with peanut allergies might want to think twice before trusting Le Chat to order take-out food.
The other new feature: “Connectors” to link services
Alongside the memories feature, Mistral also released “more than 20 secure connectors” that allow Le Chat customers to connect to business-oriented tools. These connectors, which use the Model Context Protocol (MCP), allow AI models to interact with third-party services, turning them into “agents” that can perform actual tasks.
Security risks of connectors
Although Mistral describes the connectors as “secure,” this field is still new and carries risks. A security firm recently found that 1 in 10 MCP plugins were fully exploitable, and the presence of three such plugins raises the exploitation risk to over 50%. This highlights the ongoing challenge of securing interactions between AI models and external services.
Conclusion: The delicate balance between personalization and security
The launch of the “Memories” and “Connectors” features represents an important step for Mistral AI to maintain its competitiveness in the crowded chatbot market. However, it also highlights the dual challenges facing the entire industry: how to deliver increasingly personalized and efficient experiences while ensuring user privacy and security. The 86% accuracy rate is a reminder that this technology is still under development, and users must exercise caution when entrusting vital details to its imperfect memory.
