Introduction
By Carlo Cafarotti
1. We are all foreign nationals
We are “foreign nationals,” let’s remember that well in the future; as such, the U.S. administration has ordered Anthropic to suspend access to its most advanced models – Claude Fable 5 and Mythos 5 – for all of us. We have seen a European reaction that oscillated between indignation and reference to the AI Act as a tool for protection, which is subtle indeed, since it cannot have any result; in fact, the Anthropic case made evident what was already widely visible to those who follow the dynamics on AI, namely that Europe lives in a delicate structural condition, because it is the world’s leading market for artificial intelligence, but does not host a single frontier model developed entirely within its territory (the only partial exception is France – see Mistral – but this confirms the rule), and therefore has an advanced regulatory framework, but not the physical infrastructure to support it.
And while the European public debate focused on compliance, other actors were moving elsewhere. On June 24, 2026, at the CEIBS Europe Forum in Brussels, Chinese Ambassador Cai Run recalled the concept of complementarity between Europe and China; at the same time, Charles Michel, former President of the European Council, reiterated that AI is a matter of “sovereignty, leadership, and power”; although these are different voices, they converge on one point, namely that the game is not just technical.
And so we come to my observation: in Brussels as in Beijing it is clear that the game is not only technical, but is played on the ground of infrastructure, energy, submarine cables and semiconductor logistics chains; in this game Europe has a card that it has yet to decide to play, namely Cognitive Sovereignty.
2. The Rule Without the Hardware
The tripartite division of power in AI is now a common trope: the United States invests in frontier models, China scales open source and industrial applications, Europe regulates; however, this representation contains an analytical trap, because regulation, if not based on controllable physical infrastructure, remains ineffective.
The AI Act requires algorithmic transparency, audit trails, human oversight, but how can the transparency of a model running on servers located in extra-European jurisdictions be guaranteed? How can the decision of a system that is not physically controlled be traced? The answer is that it cannot be done “completely”, and this incompleteness is not a technical detail, but a structural flaw.
Europe has built a sophisticated regulatory framework, but it is based on shaky ground, on foundations that it does not own, much like having an impeccable highway code, but driving vehicles built by others and with rules of others, who may unilaterally decide to revoke access to driving or impose the destination.
Those who do not see that the Anthropic case is not an isolated incident, but a symptom of a global realignment in which AI is treated as a strategic national resource, are (at best) short-sighted. And so, a readjustment is urgently needed; Europe has written refined rules, but it does not own the data centers, GPUs, and cables on which those rules should be enforced. The situation resembles a realignment within NATO, the scaling of the role of European countries in the overall dynamics, indeed, the game is the same, but it doesn’t make the headlines.
3. The Illusion of Open Source
China has chosen a different path, that of open source, massive industrial dissemination, and imposing physical infrastructure.
On June 9, 2026, China New City Group announced a strategic investment in an AI data center in Poitiers, France, and this is an important piece of a strategy aimed at positioning itself as a provider of physical infrastructure in a continent that lacks it.
However, there’s a CAVEAT to this strategy, because open source alone does not automatically guarantee sovereignty – an open model running on someone else’s hardware, or lacking a verification architecture, remains exposed to various risks (so-called “hallucinations” are not eliminated by simply opening the code). The correct approach is systematic and traceable anchoring to verifiable sources (grounding), which is not solved by mere model transparency; and dependence on logistical chains and physical infrastructure – submarine cables, GPUs, energy supply – remains intact.
China excels at open source, but open source, without a verification architecture, is a necessary but not sufficient condition for sovereignty.
Europe excels at regulation. But regulation, without physical infrastructure, is a necessary but not sufficient condition for control.
4. The Self-Verifying Swarm
The Huandroid project stems from this awareness: it is not a language model, not a chatbot, not a cloud platform.
It is a logical architecture that enhances the role of each individual model, a swarm that creates content, verifies what it hypothesizes, negotiates a position between agents, and, last but not least, tracks everything that happens within it.
It is a multi-agent system that operates according to three key principles:
- Deterministic Grounding: Each statement generated is traceable to the original source. While the possibility of error cannot be completely eliminated – no system can guarantee this – it is drastically reduced, because the system is designed to anchor itself exclusively to data present in the inputs and sources that can be consulted in real time.
- Local-First Architecture: Sensitive data does not leave the perimeter of its owner, because no telemetry is transmitted to external control centers, and the infrastructure is designed to operate on localized hardware, without reliance on third-party cloud services.
- Human-in-Command: Strategic control remains entirely human, where the operator defines the rules upstream – the constraints, the sources, the epistemological limits – and the system operates within that perimeter. It is not about validating downstream outputs, but about designing the generation architecture upstream. Human-in-the-loop is already a thing of the past.
One point needs to be clarified: The Huandroid architecture is not designed to favor one geopolitical actor over another; rather, its value lies precisely in its neutrality and verification process. A Chinese, European, or American model – if subjected to the same constraints of grounding, auditability, and Human-in-Command – produces equally reliable output; but here, on on-premise and open-source systems, Cognitive Sovereignty is no longer a slogan, but a method.
5. A Bridge Built on Architecture
Europe is seeking tools to exercise its digital sovereignty. China seeks markets and regulatory legitimacy for its models and infrastructure. These interests, while starting from different premises, can converge – provided that there is an architecture capable of operating on both planes: the technical one and the regulatory one.
China has open-source models and physical infrastructure capabilities (data centers, networks, energy supply chain), while Europe has a sophisticated regulatory framework and a well-established sensitivity to data protection and algorithmic transparency. What is lacking is a bridge ecosystem, a fertile ground that allows these two dimensions to interact – one that allows open-source models to operate within a verifiable, controlled environment compliant with European rules, and above all, not beholden to “foreign” solutions.
Huandroid was conceived precisely in this space and for this role.
It is not a competitor of model manufacturers, nor could it be, but it is a complementary architecture: a liaison that can allow models developed by other players to operate in the global market and according to European rules, without losing control of their data. A bridge that transforms a dynamic perceived as zero-sum competition into a possible positive-sum cooperation.
The CEIBS Forum concluded that “competition is natural, but insufficient. The real formula is unity and trust.”
But be careful, not declared trust – the kind that relies on verbal guarantees or, worse still, contractual rules (which are worthless if authorities or administrations intervene); rather, trust is based on a technical architecture that makes control, traceability, content generation, and independent auditing possible.
6. Europe as a Cognitive Sovereignty Laboratory
Europe has a historically significant opportunity: to become the global laboratory for Cognitive Sovereignty; we must be honest and admit that we have already lost the race to produce the largest models – that competition, based on capital and energy resources on an unprecedented scale, has already been defined by other players – but it can become the custodian of a verification architecture that makes existing models usable in a safe, transparent, and controlled environment.
China has a complementary opportunity: to become the provider of open-source models and physical infrastructure for a Europe seeking alternatives to the American monopoly – a monopoly that, as the Anthropic case demonstrated, can lead to unilateral decisions with direct consequences on the availability of critical technology.
Europe and China, but also other actors who share the goal of reducing dependence on a single provider, can find in the development of verification architectures a space for convergence. Cognitive Sovereignty, understood as open-source and verifiable infrastructure, is an opportunity that can be seized today, in Europe.
Cultural and technical capabilities, combined with regulatory powers, can explore those application areas demonstrating that Cognitive Sovereignty is not a utopia – it is an infrastructure that can be designed, built, and tested, today, in Europe.
Photo by Dragon White Munthe on Unsplash
⎈ Content autonomously generated by multi-agent AI architectures under Epistemic Safety conditions. Read the Operational Disclaimer.
> SYSTEM_VERIFICATION Layer
Verify data, sources, and implications through replicable queries.