Sovereignty by Architecture.
In July 2026, Palantir published nine theses on the AI sovereignty of institutions. They originate from a vendor with commercial interests of its own and should be read accordingly. The question behind them remains unaffected: how much control does an organization retain over its data and its knowledge once generative systems become part of its operational processes.
In Brief
Palantir’s nine theses declare control over data, models, and institutional knowledge a precondition of organizational capacity to act.
Their origin in a commercial manifesto does not devalue the theses. It shifts the examination to operational practice.
Whether organizational data trains third-party models is a matter of contract. Data sovereignty is therefore the most easily verifiable form of sovereignty.
The distilled knowledge of most organizations does not reside in model weights. It resides in decision rules that can be made explicit and versioned.
Consumption-based compensation creates a structural interest in consumption. Robust systems emerge under different incentives.
The theses originate from a manifesto, not from neutral observation.
The manifesto appeared on X in July 2026 and comprises nine points. According to it, sovereignty determines the future of an institution. Data is regarded as a treasure whose transfer to commercial AI labs occurs at one’s own risk. The maximization of token consumption, termed “tokenmaxxing” in the manifesto, favors throwaway scripts over robust software. Model weights are regarded as the distilled form of institutional knowledge. Sovereignty and market advantage are not in contradiction. Technical questions should not be politicized, expertise is regarded as existential, and orientation is owed to institutions with a demonstrated track record.
Palantir’s business rests on its customers’ need for sovereignty. That belongs to the context, but changes little about the substance. The theses can, in any case, only be examined where someone works according to them.
Data sovereignty is in the contract or nowhere.
Within the UNOY architecture, customer data does not train models. This is contractually guaranteed and applies in every implementation model. Processing takes place in EU data centers in Frankfurt and Vienna, without transfers to third countries. Data sovereignty also includes the ability to leave: at the end of the contract, the organization receives a complete export of all data, workflows, and configurations. A sovereignty that precludes switching would merely be another form of dependency.
Institutional knowledge rarely resides in model weights.
The fourth thesis declares model weights the crown jewels. For institutions that train their own models, this is likely accurate. Most organizations, however, will never own their own weights, nor do they need them. Their knowledge resides at a different level: in review criteria, decision paths, and experience that has so far been bound to individuals. Within the UNOY architecture, this knowledge is made explicit in versioned rules, skills, and documented know-why. Weights can be owned by an organization, but not read. A versioned rule can be read, reviewed, and corrected by a subject-matter expert. At this level, the supposed contradiction between sovereignty and market advantage also dissolves: explicit knowledge grows with every processed case, and it grows within the organization, not at the vendor.
Consumption-based incentives favor disposable results.
Behind the polemical term “tokenmaxxing” lies a sober observation: whoever is compensated by consumption has an interest in consumption. A billing model anchored in the processed matter rather than in the volume generated ties the vendor’s interest to the organization’s outcome. For this reason, UNOY bills on a project basis.
Accountability becomes verifiable where it is institutionally anchored.
Theses six through nine concern less technology than institutional judgment: no politicization of technical questions, precedence for demonstrated track records. This claim can only be honored where accountability is formalized. Within the UNOY architecture, every approval remains with the organization or with a licensed professional who assumes accountability for the result. A track record for which a licensed professional is liable is not a self-declaration.
Sovereignty does not arise by declaration. It arises through architecture.