Mistral AI has demonstrated the capabilities of its new Large 4 model, nicknamed Le Chonk, by automating the security triage, reproduction, and patching of a real-world vulnerability in the Django web framework. The process utilized 133 specialized agents to map the codebase, locate the flaw, simulate attacks, and generate a fix.
The demonstration, released on November 4, 2026, focused on vulnerabilities that Django had officially addressed in its own security updates on November 3, 2026. These included issues such as memory exhaustion and log injection, which can respectively overwhelm system resources and tamper with administrative records.
Mistral AI made the model weights publicly available on October 30, 2026, allowing users to run the software on their own infrastructure. This approach is intended to address concerns regarding the privacy of proprietary code, as companies can avoid sending sensitive data to external APIs.
The company highlights that the model maintains a refusal rate of over 96% when faced with malicious cyber prompts during benchmarking. This safety feature is designed to prevent the model from being used for offensive purposes, despite its advanced capabilities in codebase mapping and attack simulation.
The use of 133 agents represents a multi-agent orchestration strategy, where complex tasks are divided among specialized roles rather than relying on a single model. Mistral AI is positioning Large 4 as a comprehensive tool for automating the entire vulnerability lifecycle.
