Tom and Jerry in Cyberspace: Indian Researcher Sandeep Joshi Talks AI Agents, Cybersecurity, and Trusting Machines

First of all, welcome to SpacSec 2026. What drew you to attend the conference and visit ITMO?

The conference topic is extremely relevant both to current and future generations. If we look at modern trends in information security and space computing, it is clear that these fields have become enormously important for every country; today, they are an integral part of technological development and security. Therefore, we need to study these areas and actively participate in their development. This topic is also of great personal interest to me.

As for ITMO, I was aware of the university’s reputation in cybersecurity. I myself know several of the university’s professors, who are internationally acclaimed researchers with significant results in their fields. All this served as additional motivation for me; I simply couldn’t miss this opportunity.

Autonomous security systems are the subject of wide discussions. They are capable of making their own decisions on blocking or allowing certain actions without human input. How confident can we be in machines in terms of security – detecting threats and reacting to them? Where should we draw the line on automation?

I believe that automation is necessary, but it must be carried out under human control. With full automation, malicious actors may discover a vulnerability that would put the entire system at risk.

For example, automation is well suited for routine tasks: administrative work at university, the activities of individual departments, or everyday operations. But when it comes to critical systems and the security of an entire country, human involvement becomes fundamentally necessary. One wrong automated decision could potentially lead to the shutdown of an entire network. Both automation and human intelligence have their advantages and disadvantages. Therefore, I consider the optimal solution to be a combination of autonomous systems and human control.

Another hot topic are AI agents, which are already used by both defenders of information systems and their attackers. Which side is winning in this race? And why?

I think everyone is familiar with Tom and Jerry – here, the situation is similar. Malicious actors are constantly trying to attack our systems, and information security specialists must be prepared to counter these attacks. It is a constant race.

Yesterday at the conference, we discussed issues of cyber defense and offensive operations in cyberspace. In some situations, it is necessary to be prepared not only to defend but also to act preemptively. If the adversary is already nearby, it is not always right simply to wait for their attack. Sometimes it is necessary to take the initiative. But in other situations, one should focus exclusively on defensive measures. In my view, a balance between offensive and defensive approaches is necessary.

If both sides rely on AI agents, is modern cybersecurity strictly a battle of machines? Is the human role changing and is there a risk of humans disappearing from the process entirely?

If one side uses AI and the other does not, then, of course, the first side has an obvious advantage. But when both sides use AI, the effectiveness of how the technology is used becomes crucial. For example, the way intrusion detection systems are applied is very important. In my presentation at the conference, I already mentioned that, according to my observations, malicious actors use autonomous systems in approximately 70-80% of attack generation cases, while the defending side lags somewhat behind. This happens because attacks are mostly initiated with the help of AI, while defence is human-dependent: someone needs to react and make a decision, while the cost of a mistake can be high – it will have an impact on a network’s performance or even take it down. We need to learn to think like attackers. We must be able to put ourselves in the attacker’s shoes, identify weaknesses in our own systems, and build defenses based on that.

And it is important to remember that AI, like other technologies, is primarily a tool. We must not allow technology to completely escape human control. Problems arise when people become too dependent on technology and stop participating in decision-making. After all, AI was created by people. And humans must retain the leading role. We can give AI a certain degree of autonomy, but within strictly defined limits. It is necessary to clearly determine which decisions the system can make on its own and in which situations human intervention is mandatory. Ultimately, the most important decisions must remain with people.

Generative AI has significantly lowered the entrance threshold for malefactors: it can be used to create phishing texts, apply social engineering, and generate malicious code even for people with basic technical skills. How greatly has this influenced the balance between attackers and defenders? Which conventional methods have become obsolete and which ones have gained more traction?

Absolutely, both sides use AI – attackers and defenders alike. The same applies to generative AI. Today, these tools are publicly available, and almost anyone can use them. Therefore, it is important to understand that technological capabilities are available not only to information security specialists but also to malicious actors. The use of generative AI can help us temporarily limit certain types of attacks, primarily phishing, social engineering, and automated malware generation, through faster threat detection, suspicious-pattern identification, and automated responses. However, malicious actors do not disappear. They are constantly looking for new ways to attack information systems and will certainly use generative AI to develop new methods. Therefore, even if attackers encounter difficulties at some point, after some time they return with new approaches. We must constantly be prepared for this.

There are some conventional methods that are becoming obsolete like password-only authentication, manual monitoring and incident analysis, as well as static firewall rules without continuous monitoring. Other methods, on the other hand, are gaining popularity, including AI-driven behavioral and anomaly detection, zero trust architecture and continuous authentication, and automated threat hunting and incident response.

It seems that regular users find it harder and harder to tell an AI-generated fraudulent email from a real message. Are there any practical recommendations you could give on the matter? Could we teach users to detect such attacks and how effective is that?

The most important thing is raising awareness. Today, even kids can often tell that an image was AI-generated. Why do they know that? Because they are already familiar with such technologies and know about their capabilities.

The same approach should be applied to fraudulent emails. The fundamentals of cybersecurity should become an integral part of educational programs. Students need to be taught to recognize suspicious and potentially malicious content. Machine-generated content is created based on specific algorithms, which means we can detect certain recurring patterns in it. For example, if several phishing emails are generated using AI and compared with emails written by humans, it may be possible to detect recurring features. AI-generated phishing emails, for instance, may contain overly polished language, artificial urgency, generic greetings, and persuasive requests for sensitive information or immediate action. Improving digital literacy will help people recognize suspicious messages more effectively.

In your opinion, should we create dedicated educational programs for these issues? Is it enough that users learn to detect AI content on their own?

Educational programs are absolutely necessary for the most vulnerable audiences. Young people often already have a certain level of knowledge in this area. However, children and older people may require additional help. It is especially important to pay attention to the elderly, as they often become victims of fraudsters. It is difficult for them to keep up with the rapid development of technology, and they are often cut off from up-to-date information. This makes this category of users more vulnerable: malicious actors can manipulate their emotions. Therefore, the younger generation should take responsibility for improving the digital literacy of older people and educating children.

You’ve attended cybersecurity conferences all over the world – in the US, Japan, Italy, Norway, Cyprus, Morocco, and Cambodia, among others. Would you say that cybersecurity is perceived differently in different countries?

Yes, that is true. The type of cyberattack largely depends on the demands of a specific country or its geographical location. For instance, we can compare Israel and India. Israel constantly faces national security threats, while in India, our priorities are different: we are mostly focused on limiting financial fraud, embezzlement, and similar crimes. It’s impossible to create a single system that could be uniformly efficient all over the world. Different countries face different challenges and therefore require different cybersecurity methods. I see this as a fundamental principle.

In 20 years, you’ve built a career as a researcher, then head of a department and a specialized center at the Manipal Academy of Higher Education. How do you combine your scientific work, management duties, and social activities? What was the most surprising thing for you on this path?

First of all, I’ve always seen myself as a student and this approach allows me to constantly learn something new. I am a firm believer that regardless of your position, you have to share your knowledge with colleagues and students – and at the same time continue learning. When you share your knowledge, others gain new competencies and thus can take on new tasks, while you get the chance to study new fields and broaden your worldview. This knowledge exchange and lifelong learning help you tackle all kinds of professional tasks. I see this as one of the main secrets of success.

What advice would you give students who are taking their first steps in cybersecurity?

In many fields, sooner or later you will get to the point where it seems that you know everything: your work becomes routine, even boring, because there are no new challenges. In cybersecurity it’s different: here, you have to constantly keep your hand on the pulse because new challenges appear all the time. Whenever you master a technology or defence method, malicious actors are already looking for a way around it. That’s why you need to always be vigilant.

My main piece of advice is to not be afraid of new challenges. Welcome them, learn from them, see them as opportunities, and look for solutions. And, once again, keep being students, keep learning. This is one of the main principles of professional growth.

About SpacSec Conference

This year, SpacSec or International Conference on Cyber Warfare, Security & Space Computing, was held at ITMO on October 26-28. The program featured 4 plenary sessions, 59 presentations across 9 sections, and 2 roundtable discussions. The key topics were cyber confrontation (attacks on critical information systems and digital infrastructure), cybersecurity (protection of networks, systems, and data, including quantum cryptography and secure telecommunications), space computing (high-performance and quantum computing, satellite communications, and signal processing for distributed space systems), as well as related fields – artificial intelligence and machine learning, computer vision, embedded systems and IoT, and wireless ad hoc networks.

At the conference, ITMO presented an AI-based synchronous interpretation system that allowed participants to listen to talks in Russian and English in real time with minimal delays. The solution was tested by international speakers:

“I often need to interact with Russian colleagues and friends. I see that there are many incredible researchers among them who do high-quality work. However, they don’t always have it easy when they need to present their research to people speaking other languages. ITMO has now offered a solution to this problem. I congratulate the university on such an achievement and implementation of this technology and I hope that other universities will follow this example,” says Sandeep Joshi.

Interview by Nikolay Eritenko, a researcher at ITMO’s Faculty of Secure Information Technologies