Mistral Says Mistral Large 4 Coordinated 133 Agents in Django Security Demo

Mistral AI said Mistral Large 4 coordinated 133 agents in a Django security test, showing an AI-assisted code review and repair workflow without enough detail to assess the patch.

The workflow divided code examination and attack testing among agents before the model attempted a repair. The video ran at 90 times normal playback speed.

Mistral described the target as a recent Django vulnerability but did not identify the issue or publish the proposed patch. That leaves the specific flaw and repair unavailable for review.

Django published fixes for six security issues on Nov. 3: one rated moderate and five rated low. The demonstration did not specify whether it used one of those issues.

Large 4 has open weights that organizations can inspect, customize and run on their own systems. Mistral released Large 4’s weights on Oct. 29.

The video does not establish that the model found a previously unreported flaw or that its repair passed independent testing. It also gives no actual run duration, token cost or count of candidate flaws.