As AI Accelerates Cyberattacks, Cyber Defense Must Evolve Too

Today, new AI models automate attacks and demonstrate increasing capabilities in discovering and exploiting vulnerabilities, which has multiple consequences.

Automate to Regain the Upper Hand

In the face of this acceleration, defending solely at human speed is not an option.

• First, because it has become possible to orchestrate attacks with greater impact without additional skills—particularly due to the accessibility of high-performance cyber models—which lowers the barrier to entry for hackers. Consequently, the cost of a token has become a factor in both offensive and defensive cyber capabilities.
• Second, because the use of offensive AI—capable of carrying out cyberattacks autonomously—requires a rethinking of the “human-in-the-loop” concept, since human liability applies even when the machine is acting.
• Finally, because the gradual integration of agent-based AI systems into businesses has significantly increased the attack surface and associated risks.

At Orange, we believe that AI’s primary contribution to cyber defense is operational. As the volume and speed of attacks increase, detection, classification, and response must be automated. The goal is not to replace the analyst, but to enable them to address an incident much earlier, before it escalates into a major compromise.

AI can also intervene upstream. The lowering of the barrier to entry for hackers also applies to defenders, who can identify exploitable compromise vectors one step ahead. In cybersecurity, the best defense is often the one that has been planned and simulated.

AI Does Not Make the Fundamentals Obsolete

However, this race toward innovation must not make us forget that the historical principles of security remain valid. Access management, encryption, strong authentication, least privilege, and zero trust are still effective lines of defense.

However, we must innovate to address the new risks posed by agent-based AI, such as “prompt injection”—which involves manipulating AI behavior with malicious text instructions—or the hijacking of AI agents to exploit their data access privileges. These scenarios must be anticipated now.

Maintaining Human Control

At Orange, we believe that AI should be leveraged where speed, volume, and complexity exceed human capabilities, thereby serving as a defense multiplier. Analysts remain essential for setting rules and maintaining control, while AI enables the exploration of more scenarios and the effective automation of defense.

These new rules also require us to remain technology-agnostic. Depending on the situation, we must be able to choose between cutting-edge models, off-the-shelf solutions, or custom engineering—without any preconceived notions. Performance will stem from our ability to combine these different building blocks, to quickly switch models when necessary, and to avoid any critical dependencies.

In 2026, we will enter a pivotal phase. Cybersecurity professionals must incorporate a new set of skills centered on AI techniques to ensure protection and response capabilities that keep pace with evolving attacks. It is no longer possible to consider cybersecurity without AI. Defense will have to evolve at the same pace as the threat, without neglecting human oversight.