Dr. Dan Bogdanov leads Cybernetica’s Information Security Research Institute which works on technologies for governments and industry that want to secure their data, identities, assets and, more recently, AI. He leads a research team that works on international cybersecurity research projects with the U.S.’ DARPA, U.S. Office of Naval Research, European Defense Fund, Horizon Europe, European Space Agency and Estonian agencies. He is also the Head of the Standing Committee on Cybersecurity at the Estonian Academy of Sciences, where his focus is to improve the cybersecurity risk analysis quality of e-governments.
Estonia is a pioneer in digitalizing government services, and Cybernetica, the Estonian company that helped build the technology behind them, has since helped more than 35 other governments around the world. Now, as Europe worries about a possible Russian invasion and the world at large grapples with cybercrime and the escalation of military conflicts, Cybernetica is moving into defense and cybersecurity applications. In an interview with The Innovator, part of our special report on this week’s Tallinn Digital Summit, Bogdanov talks about why economic sovereignty in the age of AI may matter even more than privacy, how Cybernetica is helping European militaries move beyond the fax machine, and why governments and companies need a strategy for retaining knowledge.
Q: Estonia was a front-runner in the digitalization of government and Cybernetica has since worked with more than 35 governments. How did it start, and how did you take what you did for Estonia to the rest of the world?
DB: Not everything was invented in Estonia. When Cybernetica began its security research, we started by reading the research papers and textbooks others had written. What we did — and this is me interpreting in retrospect — was take that fresh know-how and ask: What if we didn’t have a legacy to support? What if we took the latest research, added our own innovations, and built the whole thing around it? That is what Cybernetica originally did in Estonia, and we have kept doing it.
The upside is that we have been able to surprise a lot of people who didn’t think these things could be made practical. The downside is that some of our surprises come early. Sometimes we come up with something for which there won’t be a market for 10 years, and we sit on it for 10 or, in some cases, 15 years. Innovation really is a cycle: You read, you write, you read, you write, and then you have to find the right moment when a government or an institution is willing to put it into practice and take it out into the world. For many things, that has been the Estonian government. Estonia has been a great enabler for us but as we have become bigger and more international, some of our technologies are being taken into the world by other governments or by institutions such as Eurostat and the OECD.
Q: How has your offering for governments evolved?
DB: There are different types of digital citizens and digital societies. In some countries where we have built innovations, they haven’t taken root the same way. The society wasn’t there, or the government institutions weren’t committed enough, strong enough, or resourced enough. So, there are things in our pipeline that we believe some governments will come to in five or 10 years. But there are things we have always been working on that are now super timely. Sovereignty has become timely because of global events. Some people frame sovereignty in terms of values — European fundamentals such as autonomy, privacy, and data protection. But there is also a strong economic sovereignty story, and AI makes this painfully clear.
Q: How so?
DB: If you make all your value chains dependent on somebody else’s models and inference infrastructure, then all your processes will leak value out of your country. The economy has always been global. You sell your natural resources and use the money to buy something else. That is all fine. But AI has this property that people look at it, think they understand it, and believe it can be used efficiently for everything. That is not the case. According to a recent report from Prosus,, only about 15% to 20% of AI agents earn their keep in many organizations. A lot of people are jumping into AI hoping it will make them rich, but it is not going to be like that. And while they are chasing this, they are building a dependency on another state into their education system, their legal system, and their government.
If countries stop knowing how they run, they lose the skills. It might take three generations, but at some point, ministries of education or science may find they have a totally new job: retaining knowledge relevant to the country so that some of it remains under national control, rather than coming from models somewhere else. If all knowledge is packaged into machines, the sovereignty of states will be in even greater peril than we imagine.
Q: Where does Cybernetica fit into that picture?
DB: We are not rushing to market with, say, AI agent-powered interoperability for our UXP platform. We have our vision of where it could create value, but I am currently afraid that a lot of that value might be illusory. There might be a prime minister somewhere who feels they need to let AI agents use their e-government platforms. Fair enough. I am not saying this is a bad idea. It might create excitement. But you need to look at your digital citizen. Do your citizens use personal AI agents? Will they let ChatGPT run their lives? Is there actual maturity for setting up a wider AI-enabled ecosystem in your digital government? And then you suddenly find you need other pieces, such as delegation of rights, so that if you give AI agents some rights, those rights mean something.
Estonia has become famous for suggesting that maybe we should give AI agents personal identification codes, and there is a big argument going on about that right now. Part of the initial idea was: Let’s do something and fix it later. But ‘fix it later’ might not work here, because the agent might run away with your money and buy Bitcoin for you. So, for us, the key points are that identity in the AI age must remain strong identity, transactional intent matters, and confidentiality will remain important. Who gets to see your data is something you will need to pay attention to.
Q: Can you speak about why confidentiality matters so much in the age of AI?
DB: Previously, if an unfriendly country wanted to corrupt one of your people in government or in a key industry, it needed something physical on them — compromising pictures, or some sort of charge in another country. If people grow up talking to AI agents about everything, you don’t need to do much anymore. You can manipulate people in real time by giving them personalized advice and guiding them to do things that can later be used against them.
There was a case where an AI companion service was hacked and the transcripts were leaked along with the email addresses of the people who wrote them. It was very instructive about what people talk to these machines about, and some of it was horrible. If they had said these things to actual humans, there would be grounds for criminal proceedings. Reputation is going to be a thing. Even if a society doesn’t have a concept of privacy, it may have a concept of honor and the will to protect it.
Q: Do attitudes differ from culture to culture?
DB: Very much so. I learned a lot about digital societies when I was doing standards editing at ISO. I was a co-editor of a privacy standard, and I had to integrate the rulebooks of the English-speaking countries — the U.K., the U.S., New Zealand, and Australia — with the European treatment and with Japanese concepts. I am generalizing heavily, but all three groups had totally different concepts of what we were doing. That is why I don’t think there is a single formula for a digital society. At Cybernetica, we are mindful of that. Our group includes social scientists.
Q: Can you give an example of how that plays out in your work?
DB: A recent project we did for the Estonian government is called Digital Me. It is a total reconsideration of how a person interacts with their ‘digital me’ — their data in the government. It is a consent service and a data monitor rolled into one. In that project we had social scientists and a psychologist looking at a person’s relationship with their data. Is it reasonable to expect people to manage all their consents somewhere? We don’t think so. A person can usually keep about five, plus or minus two, things in mind. Privacy laws say you can revoke consent later, which is all well and good, but I don’t even remember who I gave consent to.
Society and the economy also need tools for the mass processing of data for the common good, and that means we need to put greater protections in place. This is something we have increasingly been working on, including in projects with Statistics Estonia and European statistics bodies.
Q: Cybernetica is also moving into defense. Can you talk about that?
DB: There used to be a time when you couldn’t get a military asset across borders in Europe without sending a fax. I think this might still be the case in some places. We think this won’t do. The Estonian Ministry of Defense has been championing the Digital Military Mobility project, and Cybernetica has been leading the technical work with a big consortium of European partners. It is trying to streamline the process of moving people and assets across borders in Europe in a military setting. We are taking good, old understanding about information, data, and bureaucracy and bringing it to a place where the biggest challenge is getting many countries to talk to each other through a single technical standard. The time is right; there is demand for it. People see that efficiencies are needed, and the project is going quite well.
Q: What other defense-related work are you doing?
DB: We do a lot of cyber defense work related to defense hardware. New equipment being built, such as like armored personnel carriers and tank, is increasingly digital. It has a lot more data and may have AI components, which reads as attack surface. If I want to stop a piece of hardware from going on a mission, what easier way than to botch a software update so it doesn’t get out of the garage?
Q: You also head the Standing Committee on Cybersecurity at the Estonian Academy of Sciences. What issues are you focusing on?
DB: We bring together industry researchers, public sector officials, and university researchers to talk about the big cybersecurity threats, and using pragmatic researcher and industrial engineer insights, we look at how afraid we should be. For example, you put your stuff in a cloud somewhere. What if the data center burns or gets hit by a drone? This has been happening a lot over the last year. It is no longer a given that if you put something in the cloud, it is going to be there for sure.
Q: Data centers in the Middle East have been hit this year, wiping out capacity.
DB: Yes, and that means their services are gone. Russia and Ukraine have started hitting each other’s data centers as well. So, the question is, multi-cloud, for sure, but should one of those clouds be territorially close to you? In Estonia we have a lot of sea border. Our cables to Sweden and elsewhere are on the bottom of the sea, which is also a problem. If those cables get hit, will we still have our e-government? Some things need to continue working because society now depends on them. We have created a societal dependency on certain services. People say, ‘Oh, you’re so innovative,’ but what about the long-term sustainability, reliability, and security of these services? Can you still do things on paper?
Q: Which brings us back to the question of knowledge.
DB: Exactly. Governments need to start thinking differently about knowledge retention — and not just governments, industry too. If you use a bunch of AI agents to run your tools, your service, your product, what happens if the person who knows how the AI agents work goes away? AI agents aren’t loyal. They can be copied easily. Retaining brains and brainpower seems controversial, even illogical, when everyone says AI is coming and we don’t need to go to school anymore, but it is necessary.
Q: What are some of the other risks?
DB: This year Cybernetica completed a study for Estonia’s Ministry of Justice and Digital Affairs. People usually get scared by the easy answer — that everybody can now attack everything, that cyberattacks are democratized. That is true. And everybody is saying you must have AI agents protecting your infrastructure. But those are not the only things. We highlighted 16 risks. They include things like social services having to handle cases of AI psychosis. Another is that people won’t be able to go into a room anymore without there being 10 to 20 devices with cameras and microphones, connected to AI and always listening — smart glasses, mobile phones, smartwatches. Live translation services are very nice; it’s like the Tower of Babel in the Bible. All of this is doable today. People are still used to going into rooms where they think they can talk to each other freely. That is no longer so in some places and soon won’t be in others, because as more devices get AI capabilities, those capabilities are not always on the device. They are in the back end, meaning everything you say in some smaller rooms will be sent somewhere, and industry will control those connections.
Previously, only national security people went into clean rooms where all the mobile phones and devices went into a box before entering. We are going to have to do that at house parties soon. If businesspeople or politicians meet at a house party, they won’t be able to talk about things anymore, because somebody will have devices that are recording, automatically transcribing, and sending it somewhere. People will want personal agents with agentic memory, so their device records everything to help them do their things. Society hasn’t caught up with this yet. We are not there still, but in that study, we projected ourselves 10 to 20 years into the future and tried to imagine what is going to happen. Different societies are going to get hit differently.
For each of the 16 threats, we said what the government can do, what an organization can do, and what a citizen can do. Cybernetica is not in the business of scaring people. We are in the business of saying: ‘You want to do this? That’s fine. This is the best current knowledge on how to do it securely, so that everybody survives in the end and nobody loses their liberties, money, or health.’ The point is to have a reasonable threat model.
Q: Estonia made headlines for its AI Leap program, which puts AI tools in the hands of schoolchildren. The same team that did the AI threats study, including you, worked with the University of Tartu’s Centre for Ethics to do a risk analysis of that program. What do you see as the risks with that?
DB: If you go into the document, you will see we have a couple of concerns in there. A lot of it comes down to educational policy. AI in the classroom significantly breaks down the principles on which educational systems have been built, including the relationship between the pupil and the teacher. In addition, for Estonia, and I believe for many other countries, the AI doesn’t speak the language properly, doesn’t know the curriculum, and hasn’t read the classical literature. That means if you try to do certain courses with AI, you are going to get garbage. Then, of course, if you give all the kids these tools, a mostly foreign system will be able to drift the values of whole generations. In all the math exercises, the women will be janitors cleaning classrooms, and the men will be building race cars. I am not saying this will certainly happen. I am saying everybody on this planet is under-equipped to prevent it from happening. If you open yourself to it, you need to be very sober about it. Otherwise, you will be in a world of hurt. And what if AI goes away? Can you still teach in the classroom?
Q: What do you want governments and companies to take away from this interview?
DB: Two things. First, while you are adopting AI — and everybody is — think about a long-term strategy for retaining knowledge in your organization. It may end up being quite expensive, but if you don’t do it, you won’t be sustainable in the long run. You will not control your organization. Second, everything that can be coded is attack surface. A lot of new things are being coded with AI support. The code might be better, but many people won’t know, because it is very hard to know. If you lose access to disciplined engineers and technologists while trying to be a very digital organization, see point one. Outsourcing all your know-how to foreign AI services might, to put it very bluntly, lead to a new wave of colonialism.
