Adnkronos (Adnkronos) – Artificial intelligence is democratizing capabilities that until a few years ago were the preserve of a select few. In a matter of seconds, it's possible to cross-reference open sources, reconstruct profiles, and connect people, companies, and events. But calling all this "intelligence" risks misunderstanding. Because gathering and aggregating information isn't espionage.
This is one of the points emphasized by Marco Santarelli, an investigative analyst of information networks and a consultant on international security, terrorism, and intelligence.
In his most recent works, he has argued that the more AI influence grows, the more central Humint becomes: the gathering of information through human networks and relationships, especially when the goal is not to know what is public but to penetrate what someone wants to keep secret.
Let's start with AI. You argue that it's a leap forward compared to the first era of social networks.
With Facebook, we witnessed a gigantic collection of data: tastes, preferences, behaviors. Today, those databases already exist. With artificial intelligence, the shift, in my opinion, is from quantity to quality. It's not just about knowing what we do, but about aggregating that information, correlating it, and building increasingly sophisticated models. This is where agents and automated systems change the scale of the problem.
The process becomes much faster and much less visible to the user.”
Is it in this context that you also read the Anthropic case and the resignation of researcher Jacob Coxon?
That case struck me especially from a communications perspective. We had a researcher leaving the company warning about the risks of the most advanced systems, and almost immediately, the company's top management acknowledged the problem. Normally, faced with an internal rift of this kind, we'd expect a different dynamic. I'm not saying this demonstrates planning, but I find it interesting how the incident helped draw even more attention to the dangers of AI. In a phase like this, we need to look not only at the technology, but also at the narrative built around it.
A journalist today can ask an AI model to reconstruct a profile in seconds that would previously have taken hours or days. Isn't this losing one of the traditional informational advantages of intelligence?
"Part of that advantage is disappearing. AI's aggregation capacity makes something accessible to a wide range of people that previously required specific software, skills, and a lot of time. But we must be careful not to confuse this capability with espionage. OSINT is one thing, business intelligence or due diligence is another, and espionage and counterespionage are quite another."
"I can ask a system to reconstruct a person's profile or a company's connections. I can get a lot from open sources. But AI can't tell me if a specific person was infiltrated into an organization to steal a patent. To get that information, you still need sources, informants, a physical presence in the area, and networks built over time."
It's a distinction Santarelli also insists on in his writings: AI is joining Humint in intelligence gathering and counterintelligence activities, "not replacing" the human component.
So the more AI grows, the more important the human factor becomes? It seems like a paradox.
"For me, that's exactly right. The more technology advances, the more intelligence must maintain a human information network. AI can be a formidable tool for control, verification, and aggregation, but there's a level of information that only comes from presence on the ground. Think of a strategic port, an undersea cable, a sensitive industrial environment. We need to understand who's moving, who they're talking to, why they're there. This doesn't match the automated analysis of already available information."
There's another problem: the abundance of synthetic information makes it harder to distinguish authenticity from manipulation. "An image can be altered, a reconstruction can be biased. Intelligence agencies can't simply absorb what they find online: they must also ask themselves where that information comes from and who might have an interest in circulating it."
We come to the idea of a European intelligence agency. Is it realistic to have a Brussels intelligence agency?
"Politically, it's a good idea. Operationally, it's much more difficult. Even within individual national systems, the transfer of information follows very precise procedures, classification levels, and chains of responsibility. Imagine multiplying this problem twenty-sevenfold. The point isn't to set up an office in Brussels: it's to understand what information the national services would actually be willing to transfer to it."
Intelligence also thrives on secrets that a state considers its own. I don't think, at least in the short term, that the major European services are ready to easily surrender networks, sources, operational capabilities, or particularly sensitive information. Collaboration already exists and is very strong in various sectors. But collaboration is one thing, centralization is another.
So rather than a “European CIA” do you envision a closer network between national intelligence services?
That seems like a more concrete prospect to me, at least in the short term. There are areas where we can share more: terrorist financing, cross-border financial movements, cryptocurrencies, certain cyber threats. And there are objectives on which European interests clearly converge. But the idea that each country simply pools everything it knows is another matter entirely.
The Five Eyes model, he observes, demonstrates precisely how difficult it is to separate intelligence sharing from the political trust built over time. Santarelli recently emphasized how smaller groups of European intelligence services are already cooperating on particularly difficult targets, starting with Russia and China: it is therefore more plausible to proceed through circles of trust and joint missions than to decree the creation of a single service from above. Diplomatic Newspaper
On the cyber front, however, is the technological race favoring the attackers or the defenders?
"First and foremost, the economics of the sector are changing. The romantic image of a hacker alone in a room, using limited resources to disrupt a large organization, no longer describes the entire system. Computational power costs money, advanced security costs money, and this concentrates highly sophisticated capabilities in the hands of those with the resources to support them."
"This is why we need to address the issue of technological sovereignty. If public bodies, police forces, or sensitive structures use technologies we don't fully control, we need to know where they come from, who produces them, and what dependencies they create. In this sense, I believe the Golden Power logic should be strengthened: not just looking at a company's ownership, but at the technologies that are truly strategic for the country and their supply chain."
The same concern applies to industrial secrets: uncontrolled use of AI platforms can expose code, know-how, and confidential information, turning a corporate issue into a national security concern when sectors like defense, energy, or telecommunications are involved.
You're even proposing much stronger control over patents. Doesn't that risk becoming an overly dirigiste approach?
"I'm not talking about controlling scientific research or blocking innovation. The issue is knowing and protecting what has strategic value. Today, technology reaches both legitimate users and those who want to exploit it for criminal purposes extremely quickly. We must worry not only about how much a product costs, but also about where it lands, who can use it, and the consequences."
Ultimately, however, no technology can promise absolute security.
Absolutely. Everything connected to a network presents an attack surface. And today, cyber attacks don't just affect the computer system; they affect people. If I receive a phone call with my son's perfectly reproduced voice telling me he needs money immediately, the problem is no longer just technical: it's emotional.
This is why security must become a widespread culture. People must learn to check the origin of a file, an address, a document's metadata, and recognize anomalies. These are skills that should be taught starting in elementary school. We won't turn everyone into cyber experts, but we can make it much harder to exploit the simplest vulnerability: the human one.
In a recent article, you wrote: "AI is methodological, spying is cultural." Is this, ultimately, the synthesis?
“Yes. Artificial intelligence can process an enormous amount of information and find connections with a speed that a human could never achieve. But intelligence isn't just processing. It's understanding context, the ability to read people, build relationships, and understand what's not being said. Sometimes, while everything speeds up, the analyst's job is exactly the opposite: to slow down, separate the noise from the information, and give the decision maker something truly meaningful.”
News
webinfo@adnkronos.com (Web Info)