Mistral Large 4 deploys 133 agents to find and patch a Django vulnerability

The demonstration, released on November 4, 2026, showcased the model's ability to perform a complete security workflow on a recently patched Django vulnerability. The 133 agents coordinated to map the codebase, locate the flaw, simulate attacks, reproduce the issue, and generate a patch.

This sequence—known as triage, reproduction, and patching—represents a comprehensive approach to vulnerability management. The model, nicknamed Le Chonk, targeted CVEs that Django had addressed just one day earlier, including memory exhaustion and log injection vulnerabilities.

Security capabilities

Mistral opened the model's weights on October 30, 2026, allowing anyone to download and run it on their own hardware. The company emphasized Large 4's strengths in identifying and remediating vulnerabilities while highlighting a safety benchmark.

This refusal rate addresses concerns that the same skills used to patch bugs could be exploited maliciously. Memory exhaustion bugs can overwhelm system resources, while log injection flaws allow tampering with administrative records.

Multi-agent approach

The use of 133 specialized agents represents a shift toward multi-agent orchestration in AI security. Rather than relying on a single model for all tasks, this approach breaks down the work into specialized roles that coordinate to handle the complete vulnerability lifecycle.

Mistral is positioning Large 4 as a leading tool for automating security workflows. The open-weight release may appeal to companies wary of sending proprietary code to external APIs, potentially increasing market interest in AI-driven security technology.