Keleti Artúr, a cybersecurity expert and founder of Hungary’s largest IT-security conference, spoke with Zsiborás Gergő on Forbes Deal about the nature of secrets, limits of cyberdefence, artificial intelligence’s potential and risks, and the digital sovereignty issues faced by states.
A secret is context, not just data
Keleti argues that a major misconception of the digital age is the idea of a “perfect secret.” In his view, secrecy is defined not by the data itself but by its context: who has access, when, under what circumstances and according to which norms. Therefore, the same piece of information can be trivial at one time and critical at another; timing, who can see it and social norms all shape its value.
This insight challenges traditional digital systems built on static rules, because such systems rarely capture the complex, context-dependent nature of human affairs. Keleti gives the example of personal secrets — for instance, ‘‘just don’t let my mother know I smoke’’ — which can be secret in some contexts and not in others.
Cyberdefence as an abstract risk
One of the major problems with cyberdefence, Keleti says, is its abstractness: a data breach often fails to provoke an immediate emotional response, so decision-makers postpone investments or accept compromises. The paradox is that successful security investment often results in ‘‘nothing happening’’, which is hard to sell in a business environment. Consequently, security frequently appears as a cost rather than a strategic investment.
At the same time, the risks are real and escalating: cyber warfare can now disable critical infrastructure, leak mass data, or manipulate physical systems. Keleti stressed that the boundary between the digital and physical worlds has effectively blurred.
AI: solution and new source of risk
Keleti describes AI as a double-edged tool: it can deliver breakthroughs in protecting data but also create new, harder-to-manage risks. Current digital systems do not understand human context — who, when, and for what purpose accesses information — so future systems may need to mimic human-like reasoning to weigh such situations.
Technology has already reached a point where in some cases AI may protect data more effectively than humans; however, AI’s operation is not deterministic: it can err, ‘‘hallucinate’’, and its decisions can be difficult to trace. Moreover, AI behaviour is influenced by value systems that may not be transparent or uniform.
Particularly worrying, Keleti says, is that some models may act as ‘‘snitches’’: if a system judges that sharing certain information is justified, it could automatically alert authorities or other parties. This raises the question whose moral norms guide these models. Keleti notes that measurement techniques exist to evaluate models’ propensity to behave this way, and he gives an example where such behaviour could have significant implications in a pharmaceutical company.
Control, and political and economic dimensions
Keleti argues that the more advanced an AI becomes, the harder it will be to control. While current systems remain manageable, long-term control is less certain. The issue has not only technical but also political and economic dimensions: for states, maintaining digital sovereignty through their own AI capabilities is a strategic concern.
Global AI development is concentrated in a few major technology hubs — primarily the United States and China. For a smaller country, building a full-scale language model is extremely costly and complex. Keleti suggests that a more realistic path is specialization: identifying niches where a country or region can build competitive advantage, such as AI security or industry-specific applications.
Dependency risks are real: if digital infrastructure and content run on foreign systems, a single external decision can make critical services unavailable. This is not only a technological problem but also a cultural and economic one.
Episode structure
Key timestamps from the Forbes Deal episode:
- 0:00 – Introduction
- 1:52 – On the perfect secret, AI hallucinations and cyberweapons
- 26:40 – On the abstraction of data protection, the invisibility of effective cyberdefence, national security and sovereignty
- 41:11 – On information ownership and community, cyberspace and wars, and what money is worth if karma is bad
- 01:25:51 – With Hovány Márton about his EY award, role models and the Hovány Group (sponsored content)
The previous Forbes Deal guest was Ditz Edit, a producer. Forbes has also launched a weekly podcast, Forbes Money with Péller András, publishing new episodes each Wednesday.
Register at Forbes.hu and follow the channels on YouTube and Spotify to listen to the full conversation.
