When Machines Fight Machines: AI Is Reshaping Both Sides of Cybersecurity The next phase of cybersecurity…
Laurent H. — CEO & Founder CYBERR & HEELR
When Machines Fight Machines: AI Is Reshaping Both Sides of Cybersecurity
The next phase of cybersecurity may involve far fewer people at the keyboard. Security researchers increasingly expect future digital conflicts to be fought mainly between autonomous AI systems: one set built to break in, another built to keep them out.
Until recently, a serious cyberattack needed skilled people at every stage. Someone had to research the target, write convincing phishing messages, probe for weaknesses, move through the network, and finally steal or lock up data for ransom. Each step took time and expertise, and that limited how many attacks a criminal group could run at once.
AI agents could remove that limit. Connected systems of agents, each handling one part of the job, could carry out a whole operation from start to finish. One agent gathers information on a company and its staff. Another writes tailored social-engineering messages. A third searches for exploitable flaws, while others deploy malicious code and handle the extortion. Once such a pipeline exists, it can run quickly, around the clock, against many targets at the same time.
Defenders are expected to respond in kind. Organizations would rely less on human analysts sorting through alerts and more on groups of specialized defensive agents. These would watch networks continuously, spot unusual behavior, isolate compromised machines, and counter an intrusion in real time. The result would be a contest decided in seconds rather than days, with people mostly setting policy and reviewing outcomes rather than responding directly.
This shift raises hard questions. Fully automated defenses can make costly mistakes, such as shutting down legitimate systems or misreading normal activity as hostile. Accountability gets blurry when no person directly made a decision. And the cost of launching sophisticated attacks may fall sharply, putting capabilities once limited to well-funded groups within reach of far more actors.
What seems clear is that the balance will depend on speed and adaptability. In a world where attacks are machine-driven, organizations that still depend mainly on manual response may find themselves outpaced before they even know they’ve been targeted.
What’s the one security task you’d never fully hand over to an autonomous system?
4 reactions · 2 comments
Comments
Schomer: The one task you can never fully hand over to an autonomous system is the enforcement of core system boundaries and runtime invariants. If you let a probabilistic AI agent determine whether a line-rate transaction payload is 'safe' or 'malicious' based on a probability matrix, you are introducing a massive application-layer surface area for prompt-injection and boundary drift [Other Google Activit
Ravimurthy M.: The real challenge is not simply AI versus AI, but autonomous execution versus trusted control. As AI-driven attacks accelerate, cyber resilience must evolve beyond detection and response towards continuous validation, strong identity and access management, least-privilege enforcement, and clearly defined boundaries for autonomous decision-making. In OT/ICS and other safety-critical environments,