Orange Cyberdefense announces Mistral AI partnership for security services
Orange plans to introduce a cybersecurity variant of Mistral’s AI model gradually, starting with selected services for corporate and public-sector clients.
Orange Cyberdefense, the French telecom group’s digital security arm, announced a partnership with Mistral AI on 29 October 2026 to bring a cybersecurity variant of Mistral’s latest model into selected services for corporate and public-sector clients, according to Reuters. The planned use could change how the company tests defenses and investigates breaches, although results have yet to be demonstrated.
Orange said the rollout would be gradual and that it would use computing capacity located in Europe to run models dedicated to its cybersecurity operations. The first uses are intended to identify weaknesses in clients’ defenses before attackers find them and to help investigate breaches faster. The company did not give a completion date for the rollout.
What the partnership means for Orange’s clients
The initial scope is limited to selected services, rather than every product Orange Cyberdefense provides. Its customers include businesses and public-sector bodies, making the proposed tools relevant to organizations that need to assess vulnerabilities and respond to incidents. Reuters described Orange Cyberdefense as one of Europe’s largest cybersecurity providers and reported that it generated €1.3 billion in revenue in 2025. Financial terms of the Mistral agreement were not disclosed.
Mistral’s model and the risks of open access
The deal comes as cybersecurity providers confront attackers using AI to increase the scale of attacks, Reuters reported. Separately, WIRED reported that Mistral calls its finalized Le Chonk model the world’s best open-weight model for automated coding and cybersecurity. That ranking is Mistral’s own assessment. The company said it refined the model with increasingly difficult coding and cybersecurity problems; WIRED reported that the finalized version is freely available.
Open-weight models can be downloaded and customized, giving defenders access to tools they can run themselves while also making those tools available to potential attackers. Rafe Pilling, a Sophos threat-intelligence executive, told WIRED that downloadable models can be abused as well as used productively. He said monitoring and controls matter regardless of where a model originates, because advanced models require supervision.
Orange’s stated first applications address two stages of security work: finding a vulnerability before an attack and examining a breach afterward. The partnership announcement sets out those intended uses and a limited starting scope; it does not establish how much faster investigations will become or whether clients’ defenses will improve once the tools are deployed.
Read in depth: Business
Volkswagen reports €5.8 billion third-quarter loss as Porsche charge and China slump weigh
The German carmaker's July-to-September loss widened sharply from a year earlier, with a large Porsche writedown and falling deliveries in China adding to pressure on its results.
World Bank forecasts 23% commodity price rise as energy and shipping shocks mount
The bank projects higher energy, metals and food prices in 2026, with further risks from conflict, disrupted grain shipments and El Niño.
Sources and context
- Orange unit signs deal to Mistral's latest AI cybersecurity modelReuters, republished by CNA
- Mistral Says Its AI Model ‘Le Chonk’ Has Leapfrogged China in the Open-Weight RaceWIRED
AI-assisted article checked against the listed sources. NewsJaws did not conduct interviews or attend the reported events.