Orange Cyberdefense and Mistral AI Team Up on AI-Assisted Penetration Testing and Incident Response

Orange Cyberdefense, the cybersecurity arm of the Orange group, announced a strategic partnership with Mistral AI on October 29 to build cybersecurity services powered by artificial intelligence, La Tribune and BFM TV reported. The two French companies will develop AI agents designed to find security flaws before attackers do and to help organisations respond once an attack has happened.

Under the agreement, Orange Cyberdefense gets early access to a version of Mistral Large 4 that has been optimised for cybersecurity work. The first two services cover AI-assisted penetration testing, where specialists probe an organisation’s systems to uncover weaknesses, and post-incident investigation and response, where the models speed up the analysis of digital traces left by an intrusion. The partners say critical actions will remain under the direct responsibility of human experts, with the AI used to speed up and extend their work rather than replace it.

The deal is a response to a shift in the market. Since the spring, American labs have released models aimed at security work, notably Anthropic’s Claude Mythos and OpenAI’s GPT-5.6-Cyber, and testers have reported finding vulnerabilities at a scale that was previously out of reach. That has pushed European defenders to look for capable tools they can run under their own control.

What it means in practice

The partnership builds on Mistral’s momentum since it unveiled Mistral Large 4. Euronews reported that Mistral released downloadable weights for the model on the same day, saying its score on the Terminal-Bench 4.0 coding test rose from 28 to 41 per cent and that it uses 47 per cent fewer tokens than China’s GLM-5.3. European bodies are also paying attention, as when the EU cybersecurity agency tested Mistral models for vulnerability work, and banks are moving too, with Crédit Agricole bringing Mistral in with cybersecurity as a core track.

For clients, the appeal is a European supplier pairing a French model with a large team of security consultants. The open questions are how well the cyber-tuned model performs against its American rivals in real engagements, how clients’ data will be handled during testing, and what safeguards stop such tools from helping attackers as well as defenders.

What to watch next is when the first services reach Orange Cyberdefense customers, whether the cyber variant of Mistral Large 4 is offered more widely, and how French and European regulators treat offensive-capable AI in security work.