Protected: Assessing Sovereign AI: A Two-Pronged Framework
CSET Georgetown Emily Tavenner
CSET rolled out a framework for judging countries' "sovereign AI" bids, but the report itself sits behind a locked page. All that surfaced publicly was a CSET researcher's take on Intel's chip comeback in the WSJ.
Georgetown's Center for Security and Emerging Technology teased a new framework this week called "Assessing Sovereign AI: A Two-Pronged Framework." The title alone signals ambition: a structured way to evaluate how nations are trying to build independent AI capacity, whether that means owning their compute, controlling their data, or fielding homegrown models instead of renting capability from American or Chinese labs. But the actual paper is walled off, marked protected, unreachable to anyone outside whatever access list CSET has drawn up. So the substance of the two prongs — what they measure, which countries score well, what counts as genuine sovereignty versus branding exercise — stays a mystery for now.
What did surface, almost as a footnote on CSET's site, was a mention of researcher Jacob Feldgoise contributing to a Wall Street Journal piece on the U.S. government's growing hand in semiconductor manufacturing. That article looked at Intel's attempt to claw its way back to the top tier of American chipmakers, at a moment when Washington has been throwing money and policy weight behind domestic fabs. It's a smaller story than a sweeping sovereignty framework, but it's arguably a more concrete one: chips are the actual substrate sovereign AI ambitions rest on, and Intel's fortunes are a decent proxy for whether the U.S. can back its own rhetoric with silicon.
There's an obvious tension worth sitting with. CSET, an academic-adjacent think tank that positions itself as a source of public policy insight, is choosing to gate a piece of research whose entire subject is national self-reliance and transparency in AI development. Meanwhile the only visible output tied to that research effort is a comment about government intervention in chip supply chains, delivered secondhand through a major newspaper rather than through CSET's own channels. It's a strange split-screen: sovereignty theorized in private, sovereignty practiced in public through subsidies and factory announcements.
Until the framework itself becomes accessible, anyone trying to gauge which countries are actually pulling off sovereign AI, versus just talking about it, is stuck reading between the lines of adjacent coverage like the Intel piece. That's not a knock on the researchers; access decisions get made for all sorts of reasons. But it does mean the more interesting half of this story, the actual criteria for judging sovereign AI claims, remains locked up exactly when governments worldwide are making those claims loudest.
My take
Locking a report about sovereign AI behind a password is the kind of irony that writes itself, and it says something about how policy research on this topic tends to work: lots of framing, not much public accountability for the framing itself. The Intel angle is the part that actually matters — fabs, subsidies, and who controls the compute are where sovereignty gets decided, not in a two-pronged rubric nobody outside a mailing list can read. Anyone serious about tracking sovereign AI should watch chip investment numbers and export controls, not white paper titles.
Read more about this at: CSET Georgetown
