AI Is Changing Warfare. Europe Must Not Outsource the Decision Layer.

The Bottom Line Upfront

Artificial intelligence is already changing defense, cyber operations, intelligence analysis, and critical infrastructure protection. Europe cannot answer this shift by renting black-box decision systems from foreign vendors. The strategic asset is not the model alone. It is the sovereign decision layer: the ability to process sensitive data locally, inspect every recommendation, preserve evidence trails, enforce policy inside the secure boundary, and keep human accountability meaningful. Lascaris is built around that principle. It supports AI-assisted intelligence and cyber defense workflows without forcing agencies or critical infrastructure operators to surrender control of data, reasoning, memory, or auditability.

‍

The Model Is Not the Strategic Asset. The Decision Layer Is.

Artificial intelligence is no longer a distant defense topic. It is already entering cyber operations, intelligence analysis, drone coordination, open-source intelligence, fraud detection, border security, and critical infrastructure protection. The speed of conflict is changing because the speed of information has changed.

European security teams now face machine-speed events. A coordinated cyberattack against a transport operator, energy provider, water utility, or government network does not wait for a weekly intelligence briefing. It unfolds across logs, alerts, public channels, leaked credentials, network telemetry, satellite images, supplier systems, and social media narratives at the same time.

Human judgment still matters. It matters more than ever. But human operators can no longer manually process every signal before the situation changes. The old Observe, Orient, Decide, Act loop is under pressure. Observation has become continuous. Orientation is now data-intensive. Decisions are made under severe time compression. Action must be governed, logged, and reversible where possible.

This is why AI is moving into defense and national security. Not because governments should automate judgment. Not because machines should replace commanders, analysts, or public authorities. AI is entering the field because the volume and speed of modern threats have exceeded traditional human-only workflows.

The United Kingdom National Cyber Security Centre has warned that AI will affect the effectiveness of cyber operations over the near term. Harvard’s Ash Center has also described weaponized AI as a serious threat to national security, democratic institutions, cybercrime prevention, and public order.

These warnings should matter to European governments because they point to the same operational reality: adversaries will use AI to compress time, lower cost, and scale attacks.

Europe cannot answer that reality with policy language alone. It needs sovereign systems that can help operators understand, verify, and act without surrendering control of the decision layer.

‍

The Trap of Outsourced Cognition

Many institutions understand data sovereignty. Fewer understand cognitive sovereignty.

Data sovereignty asks where the data is stored, who can access it, and which laws apply. Cognitive sovereignty asks a harder question: who controls the reasoning process that turns data into operational decisions?

A defense ministry may keep data inside a European data center and still outsource the actual decision layer to a closed foreign platform. A national cyber agency may protect its telemetry but still rely on a proprietary ontology, remote model updates, hidden retrieval logic, or vendor-controlled workflows. A critical infrastructure operator may satisfy residency requirements while its most sensitive incident triage depends on systems it cannot fully inspect.

That is not sovereignty. That is local storage with outsourced cognition.

This distinction matters because AI decision support systems are not passive dashboards. They shape what operators see first. They rank alerts. They correlate entities. They recommend actions. They summarize uncertainty. They decide which evidence appears important and which evidence disappears into the background.

When that process is closed, the human operator is not fully in control. The operator may still approve the final decision, but the system has already shaped the field of possible conclusions.

This is the weakness in many commercial defense AI arguments. Palantir’s article on ethical AI in defense decision support systems makes several valid points. It discusses operational reality, legal constraints, audit trails, and human responsibility. That discussion is useful. But for European defense and public-sector operators, it is not enough.

Ethics in defense AI cannot depend on a vendor promise. Auditability cannot be a feature description. Human oversight cannot be reduced to a checkbox in a procurement document.

For national security, these properties must be enforced by architecture.

‍

Human-in-the-Loop Is Not Enough

“Human-in-the-loop” sounds responsible. In many cases, it is necessary. But it is not sufficient.

A human cannot meaningfully supervise a system they cannot inspect. An analyst cannot challenge an AI-generated recommendation if the evidence chain is hidden. A commander cannot evaluate a decision support output if the retrieval path, data lineage, model version, policy rules, and confidence boundaries are locked inside a vendor-controlled platform.

In that situation, human oversight becomes theatre. The machine narrows the decision. The human carries the responsibility.

This is especially dangerous in defense, intelligence, and critical infrastructure environments because the cost of error is not abstract. A false correlation can escalate a cyber incident. A manipulated data source can distort an intelligence picture. A bad recommendation can redirect scarce response teams during a crisis. A silent model update can change system behavior without a national authority understanding what changed.

The problem is not AI itself. The problem is uninspectable AI inside operational systems where accountability must remain public, legal, and sovereign.

European agencies should not reject AI. That would be naive. They should reject black-box dependency. They should reject architectures where critical reasoning leaves the jurisdictional boundary. They should reject systems where audit logs are controlled by the same vendor whose outputs must be audited.

The right question is not: should AI support defense decisions?

The right question is: can every AI-supported decision be traced, challenged, reproduced, and governed inside the sovereign environment?

‍

Critical Infrastructure Is the Immediate Battlefield

The most urgent use case is not science fiction warfare. It is critical infrastructure resilience.

Transport, water, energy, healthcare, telecommunications, logistics, and government services are already targets. Cyberattacks against these systems can create physical consequences without a conventional military strike. A disabled logistics system can affect emergency response. A compromised water utility can trigger public safety risks. A ransomware attack against healthcare can delay treatment. A disinformation campaign can amplify panic during a real incident.

UK Parliament POST has written about artificial intelligence for cyber resilience, including the use of AI to identify unusual activity, detect fraud patterns, flag AI-generated phishing, summarize incidents, recommend responses, and assess third-party risks.

That is exactly the class of work where AI can help. But it also shows why architecture matters.

A critical infrastructure operator cannot rely on an AI system that only works when connected to a foreign cloud. A national incident team cannot wait for remote reasoning when networks are degraded. A government agency cannot expose sensitive telemetry to a frontier model it does not control. A European public authority cannot build resilience on infrastructure that may be changed, withdrawn, or legally compelled outside its own jurisdiction.

For critical infrastructure, sovereign AI means local continuity. The system must keep working when connectivity is constrained. It must process sensitive data where that data is produced. It must preserve evidence. It must support operators without replacing their legal responsibility.

This is where Europe needs a different architecture.

‍

From Centralized Platforms to Sovereign Decision Fabrics

The next generation of defense AI should not be built around one central platform that absorbs all data, all memory, and all decisions.

It should be built as a sovereign decision fabric.

In this architecture, the event stream becomes the operational source of truth. Every sensor reading, alert, intelligence report, analyst note, system state, and AI recommendation is treated as an event. These events are recorded, ordered, governed, and retained. The system does not hide reasoning inside an opaque application state. It emits decisions and supporting evidence back into the fabric as traceable records.

That changes the trust model.

Instead of asking a vendor to promise that its platform is ethical, the institution can inspect how the system behaves. Instead of trusting a model summary, the operator can see the evidence path. Instead of relying on an external platform to respect local rules, policy enforcement happens inside the secure boundary. Instead of moving sensitive data into a central black box, AI agents act where the data already lives.

This is not just a technical preference. It is a sovereignty requirement.

A sovereign decision fabric should have five properties.

First, data locality. Sensitive operational data should remain inside the environment where it is governed.

Second, local reasoning. AI inference, retrieval, graph memory, and policy checks should run inside the sovereign boundary, including edge or air-gapped deployments where needed.

Third, immutable evidence trails. Every recommendation should be linked to the data, model, rules, and context that produced it.

Fourth, inspectable policy enforcement. Rules of engagement, legal constraints, compliance policies, and escalation paths should be visible and testable.

Fifth, operational continuity. The system must support degraded networks, local autonomy, and controlled synchronization without depending on a remote cloud brain.

This is the difference between buying an AI application and owning a decision layer.

‍

Where Lascaris Fits

Lascaris is built for this shift.

It is not another foreign-controlled black-box platform. It is a sovereign intelligence framework for high-assurance environments where data control, reasoning control, and auditability are not optional.

Lascaris is designed for European defense-adjacent, government, cyber resilience, and critical infrastructure operators that need AI-assisted decisions without surrendering operational cognition. It supports intelligence triage, incident analysis, OSINT fusion, evidence correlation, cyber defense workflows, infrastructure monitoring, and decision support under local control.

‍

The architectural behind is simple: data stays in place, compute comes to the data, and every AI-supported action must leave a trace.

Lascaris builds on the same principles behind Streaming Intelligence, Federated Intelligence, and the Scalytics private AI platform. Event streams provide the operational backbone. Federated execution keeps data movement under control. Local models and controlled agents support reasoning inside the secure environment. Audit logs and lineage preserve accountability. Policy enforcement remains part of the architecture, not a slide in a compliance deck.

For European public institutions, this matters because the strategic asset is not the model alone. Models will change. Vendors will change. Regulations will change. The durable asset is the sovereign decision layer: the ability to observe, reason, decide, and audit under your own authority.

‍

Europe Needs Control, Not Dependency

Europe should not copy the American defense AI market. The strategic incentives are different. The legal frameworks are different. The sovereignty requirements are different. The risk tolerance is different.

European defense, intelligence, and public-sector operators need systems that respect jurisdictional boundaries by design. They need AI that can support human decision-making without absorbing it. They need resilience across critical infrastructure, not demos that only work in clean cloud environments. They need auditability that survives procurement cycles, political scrutiny, and post-incident investigation.

That means Europe must move beyond the shallow version of digital sovereignty.

It is not enough to say the data is hosted locally. It is not enough to say the vendor has an office in Europe. It is not enough to say a human approves the final output. If the reasoning engine, memory layer, update path, and audit trail are controlled elsewhere, the decision layer is not sovereign.

AI will change warfare. It will also change cyber defense, intelligence work, emergency response, and infrastructure resilience. The countries and institutions that benefit will not simply be those with the biggest models. They will be the ones that control how models are connected to evidence, policy, operators, and action.

The future defense question is not whether AI belongs in national security. It already does.

The real question is whether Europe will own the systems that shape its decisions.

For sovereign agencies and critical infrastructure operators, the answer should be clear: do not outsource the decision layer.