AI Accelerates Both Attacks and Defence

Advanced AI models have quickly changed how vulnerabilities are discovered, and attacks are assembled. Security teams can no longer rely on the old pace of analysis because work that once required several people can now be automated with agents. ReversingLabs is therefore developing agent-based approaches to security operations, while warning that the expertise needed to manage such systems will become even more critical.

The rise of generative and agentic AI in cybersecurity can no longer be treated as just another technology trend. Over the past eighteen months, newer models have become increasingly capable of connecting steps, analysing software and carrying out more complex tasks. For attackers and defenders alike, this changes the operating tempo. Igor Lasić, Senior Vice President of Technology at ReversingLabs, says the industry is under pressure because analytics, open-source package inspection and vulnerability discovery are all moving faster than before.

“A huge revolution has happened over the past eighteen months,” Igor Lasić tells ICTbusiness Media – ICTbusiness.info & ICTbusiness TV. What matters most, he argues, is that the newest models are no longer merely assisting analysts. They can assemble sequences of actions and attempt to exploit weaknesses. Until recently, such claims often sounded like marketing, but Lasić says current practice is proving otherwise. Models can work faster than people and complete certain operations in minutes, while a human team would need much longer.

That speed changes the risk calculation. It does not necessarily mean that the industry is facing an immediate apocalypse, but attacks may become more intense, more frequent, and more automated. Lasić is not trying to create panic. His point is that AI tools are no longer used only for experimentation. If models can connect several steps, identify a weakness and attempt to exploit it, defenders need systems that are equally fast and automated. “I expect attacks to become more aggressive,” he says.

The software supply chain is especially exposed. Open-source packages, dependencies, libraries, and components have long been a major source of risk in modern applications. AI accelerates that risk because models can help examine large volumes of code, correlate information from many sources, and surface weaknesses faster. Software teams are therefore under increasing pressure to identify and fix vulnerabilities quickly, while also understanding how AI is being used by attackers.

ReversingLabs is responding by developing its own approach to security operations, in which agents can take over tasks previously handled across several layers of analysts. Lasić describes a security operations centre where the first layer monitors alerts, the next performs deeper analysis, and more specialised experts work on protection and attribution. “Our view is that we can replace most of those layers with our agents,” he says. That does not mean removing people from cybersecurity, but shifting their role towards supervision, interpretation and decision-making.

The harder issue may be expertise. A single experienced expert working with agents can initiate work that once required a team of ten or fifteen people. That improves productivity, but it raises a new question: how will the industry train new experts if younger professionals no longer pass through the same operational learning layers? “Getting to that level of expertise will be a problem,” Lasić warns. In cybersecurity, knowledge is built not only through documentation, but through incidents, mistakes, investigations, and response work.

AI therefore creates both acceleration and a new skills gap. For attackers, it lowers the cost of attempting more actions. For defenders, it offers a tool for faster reaction. Models are becoming better at planning and writing steps that even experienced specialists would not always formulate as quickly. Lasić expects an evolution in which human work increasingly relies on agents, while the real differentiators will be domain expertise, data quality and disciplined decision-making.

The security industry cannot expect a return to the previous pace. Automated attacks and agent-based defence are becoming part of the same technological cycle. ReversingLabs is building systems intended to help people see threats earlier, understand their context and launch protection faster. AI will not eliminate the need for specialists, but it will change what specialists do. Those who know how to manage agents, check their outputs, and connect technical signals with real-world risk will play an increasingly important role in protecting software ecosystems.