A framework document from Palantir, subtitled "15 steps that every government and company can take to compound their alpha in the age of AI."
The fifteen steps
| Layer | Steps |
|---|---|
| Foundations | Ensure zero data retention (ZDR); determine your AI decision tree; identify opportunities in your architecture |
| Model layer | Guard against misaligned incentives; maximize model liquidity; own the model flywheel |
| Compute layer | Decide hardware based on assurance; own adaptable hardware for sensitive workflows; verify compute you don't own |
| Control layer | Be model agnostic; implement granular permissions; audit and log; practice adaptive cybersecurity; build by branching; own the context flywheel |
Reading the framework
Two design commitments run through it. Model liquidity and agnosticism — the ability to switch providers without re-architecting — is the framework's answer to vendor dependence, and it is the same diagnosis the Stanford HAI sovereignty brief reaches from the buyer's side when it finds that commercial offerings "reconfigure, rather than eliminate, dependence." Owning the context flywheel locates durable advantage in the accumulated organizational data and workflow context rather than in the model, which follows from treating models as substitutable.
The document is published by a vendor whose products implement the architecture it recommends, and should be read as an argued position rather than a neutral assessment. Its framing of the goal — to "compound their alpha" — is competitive rather than protective, distinguishing it from the national-sovereignty literature that shares its vocabulary.
Relationships
- related: The Commercial Landscape of AI Sovereignty Offerings (Stanford HAI, July 2026) — reaches a compatible diagnosis of vendor dependence from the purchaser's perspective
- supports: Model Liquidity — where the term is set out
- related: AI Sovereignty, Sovereign AI, Palantir Technologies, Sovereign AI (Product Concept)