Cybersecurity and AI: how can businesses prepare?

Last October, numerous government agencies and organizations across Europe had to urgently take their websites offline after two vulnerabilities were discovered in PHP, a programming language widely used around the world. These vulnerabilities, which had been present in the code for more than ten years and could allow unauthenticated remote code execution, were discovered with the help of “frontier models.” This incident marks the beginning of a new era in which AI makes it possible to identify complex vulnerabilities beyond human capabilities.

These highly capable frontier models can accelerate the identification and analysis of vulnerabilities, and potentially their exploitation. Attacks can now be carried out at unprecedented speed and scale, raising critical cybersecurity questions for businesses. How can they prepare for these changes? How can they protect their assets without giving in to panic?

At Orange, we believe the answer lies in a methodical, considered approach grounded in trust and expertise.

Frontier models’ ability to process vast amounts of data could lead to an exceptional increase in the volume of attacks and vulnerabilities. Today, a company such as TotalEnergies blocks nearly one million cyberattacks a month. Frontier models will significantly increase that number, creating serious challenges for security teams.

At the same time, attacks are getting faster. The time between the discovery of a vulnerability and its exploitation is shrinking. Between 2018 and 2026, it fell from two years to one day (source: zerodayclock.com). Frontier models will continue to shorten this window while also creating increasingly sophisticated new attack paths. Whereas businesses once had weeks to fix a vulnerability, they now need to act within days or even hours.

Does this acceleration mean businesses need to rethink their entire strategy? At Orange Business, we believe it calls instead for faster, more effective implementation of cybersecurity principles that have already proven their worth: businesses need to know what to protect, secure their infrastructure, and be ready to detect, respond and remediate.

AI makes these cybersecurity fundamentals more effective, but also more essential. That is why we need to strengthen each of them with AI, enabling us to act more effectively and at machine speed.

Our partnership with Mistral is part of this approach. It gives Orange Cyberdefense early access to a version of Mistral Large 4, optimized for advanced cybersecurity workflows. This means we will enhance our security products and services with Mistral technology. We will begin by integrating it into penetration testing and technical audits, allowing us to explore a wider range of attack scenarios, as well as into incident investigation and response, helping accelerate log analysis and attack reconstruction.

In response to the challenge posed by the rise of frontier models, our belief is that we need to strengthen the fundamentals, accelerate execution and use automation, all under human supervision.

First, businesses need to anticipate and prepare. As AI becomes more autonomous, visibility and control become increasingly essential. “If an AI agent makes an unexpected decision tomorrow, will we be able to reconstruct exactly what it did, why it did it and what impact it had?” This is the central question businesses need to be able to answer, as the Hugging Face incident in July 2026 highlighted.

This means maintaining continuous visibility into exposed assets, deploying observability solutions, and training and supporting teams so they can assess critical vulnerabilities much more quickly.

Next comes infrastructure protection: strengthening security across LAN, WAN, MPLS and 4G/5G networks; testing defenses; reducing the attack surface by limiting unnecessary external access; and isolating critical systems.

Finally, businesses need to use AI to improve the speed of detection and response.

Because this is first and foremost a cybersecurity challenge, expertise in the field is key to quickly translating risk into priorities and concrete actions.

Orange provides support spanning cybersecurity, private network management, cloud environments and AI-powered infrastructure, all essential areas for addressing the challenges posed by frontier AI models. Our expertise has been proven at scale: we operate 13 CyberSOCs worldwide and are the only European provider to own and control our global backbone. In France, Orange Business operates 450,000 routers and has already completed more than 450,000 operating system updates related to vulnerabilities in 2026.

We apply to ourselves the solutions, security practices and operating models we offer our customers. For example, we test the latest frontier models in our own AI lab. And because speed is key, we have enhanced our SOC’s capabilities with AI, cutting investigation time per incident in half through our partnership with French startup Qevlar.

Finally, our strategic AI partnerships give us in-depth knowledge of new models and their ongoing development, both to make the most of their capabilities for our customers and to develop appropriate security responses.

The rise of frontier AI models does not make cybersecurity principles obsolete; it makes their implementation more urgent. Businesses that improve visibility, strengthen their infrastructure and accelerate their response capabilities will be better equipped to face constantly evolving threats.