Orange Cyberdefense said Thursday it had gained early access to a cybersecurity-focused variant of Mistral Large 4 for security testing, as the companies plan gradual AI use in selected customer services.
Orange Cyberdefense, a cybersecurity business within the French Orange group, said it had become Mistral AI’s first European managed security services provider design partner. Initial work will cover penetration testing, technical audits and investigations after security incidents. Orange Cyberdefense said feedback from its specialists would help improve the variant for those tasks.
Hugues Foulon, CEO of Orange Cyberdefense, described the AI-era cybersecurity challenge as “volume, speed, trust.”
Orange Cyberdefense said its evaluation would use access management, activity traceability and model security checks. Its specialists will retain responsibility for critical actions. Mistral AI says it screens enterprise partners and contractually limits testing to controlled assessments of systems they own or have permission to assess.
Orange Cyberdefense plans to use European computing resources, and Mistral AI says its own data centers in the region will run that capacity. The companies linked European infrastructure to maintaining control over AI used in sensitive security work.
Mistral describes the underlying Large 4 model as having one trillion total parameters and 52 billion active parameters. The company says it trained the model using 3,800 NVIDIA Grace Blackwell GPUs at its European data centers. Large 4 entered public preview on Oct. 6, and Mistral has since made its model weights available for users to run on their own systems. The cybersecurity-focused version remains in preview for enterprise design partners. Mistral has also said its evaluations identified a risk that the model could assist real attacks and that it is revising safeguards.
Orange Cyberdefense reported €1.3 billion in revenue in 2025. The companies did not provide a production rollout date or contract value in their announcements.
