Natalia Kaspersky: We Are Seeing the Emergence of a New Market for AI System Security

AI agents and multi-agent systems are becoming a new corporate standard. They are being given access to vast amounts of sensitive information and entrusted with management decisions. Can digital security be maintained as these systems become so deeply integrated into our lives and work? And is there a risk of machines slipping out of human control? These questions are on many people’s minds today. Natalia Kaspersky, president of the InfoWatch group of companies, which develops solutions to protect corporate information against leaks and industrial systems against intrusions, shared her views on the future with the IT Russia media platform. One of the most prominent experts in information security, she founded Kaspersky Lab and led it for 10 years. She has since headed the InfoWatch group of companies for many years and serves as chair of the board of the Association of Software Developers (ARPP), known as Otechestvenny Soft (Domestic Software).

– Natalia Ivanovna, which areas of advanced technology do you consider the most promising?

– Everyone is talking about generative artificial intelligence right now. It is at the peak of its popularity. But I have noticed that another topic is coming to the fore this year. Both businesses and governments have finally recognized that AI entails certain risks. We are already seeing the emergence of a new market focused on securing AI systems. Demand is growing for LLM firewalls (a software layer between a user and a neural network designed to prevent instruction bypasses and data leaks – IT Russia note), systems for auditing AI agent activity, cybersecurity solutions, and various types of analytics. This market will clearly continue to grow because there is a need for these products.

– Are there any mandatory security protocols today that can protect companies from serious risks when working with advanced technologies?

– Before implementing a security system, any company develops what is known as a threat model, identifying the avenues through which it could be attacked and the assets it needs to protect as a priority. For a bakery, for example, those assets might be its production facilities; for an IT company, its intellectual property. The security tools are chosen accordingly, and the protocols will differ. In any case, however, when using artificial intelligence, companies need to have safeguards in place to prevent leaks of sensitive information. Some approaches are already available, but security issues have not been fully resolved. There are still relatively few solutions, and most are at an early stage. In my view, these are precisely the solutions that will shape the future.

– What do you see as the future of artificial intelligence itself?

– You know, the Chinese say that making predictions is a fool’s errand, especially when it comes to the future. The Turing test was passed in 2023 (ChatGPT reached a level at which it became difficult to distinguish it from a human in conversation – IT Russia note). Large foundational models followed neural networks, then came AI agents, and now multi-agent systems are being actively deployed. Many people are talking about so-called superintelligence, or super-smart AI, that will think like a human. Frankly, I am not entirely convinced. First, it is unclear why we would need something like that. Second, the technology underpinning AI today is, by and large, a process of mixing data and producing probabilistic, and therefore not necessarily accurate, answers.

We take a huge dataset, mix the data together, and calculate various correlations, which we then use to draw conclusions. It is already clear that this technology has limitations. I can imagine that another breakthrough will come at some point, taking us to the next technological level. But because I work in information security, I would like us to recognize the risks posed by new technologies as we advance and learn how to defend against them. These technologies are penetrating deeply into our systems and can reach things that were previously beyond their grasp, such as vulnerabilities in software systems.

For example, we applied a certain model to our own software and found 185 new vulnerabilities. And it is still unclear what to do about them. Information security now requires extraordinary effort because any piece of software may contain a huge number of vulnerabilities that attackers can exploit.

In short, we are entering a new paradigm, and humanity has yet to fully grasp all the risks it entails. We must understand those risks before rushing headlong into a bright and wonderful future. This is especially important given the actions of countries that are hostile toward us. That is precisely why our government places so many restrictions on the use of artificial intelligence. Some of those restrictions may be clumsy, but the objective is clear. If we do not investigate these issues ourselves, our adversaries will examine our software, our systems, everything. And once they do, all our systems could be left completely defenseless. We cannot allow that to happen.