black O2, inc.
ACTIVE
    FOR-001 / 70_LiteratureLIT-007

    Computing Power and the Governance of Artificial Intelligence

    Verified literature

    title
    Computing Power and the Governance of Artificial Intelligence
    Type
    literature
    Identifier
    LIT-007
    Citation
    CIT-017
    authors
    • Girish Sastry
    • Lennart Heim
    • Haydn Belfield
    • Markus Anderljung
    • Miles Brundage
    • Julian Hazell
    • Cullen O'Keefe
    • Gillian K. Hadfield
    • Richard Ngo
    • and 10 further authors
    year
    2024
    venue
    arXiv preprint (13 February 2024)
    identifier
    arXiv:2402.08797
    url
    https://arxiv.org/abs/2402.08797
    retrieved
    2026-11-13
    Textual authority
    B
    Corroboration
    4
    Status
    Verified
    bears on
    • [[SYS-003]]
    • [[CON-009]]
    • [[INS-007]]
    • [[CNF-007]]
    contests
      Tags
      • literature
      • compute-governance
      • institutions
      • policy
      • verified

      Literature Record: Computing Power and the Governance of AI

      What It Establishes

      Compute is the chokepoint that governments and firms have actually begun to govern: investing in domestic capacity, restricting flows to competitor states, subsidising access to favoured sectors. The paper's argument is that these measures "only scratch the surface," and it lays out compute as a governance lever across the development and deployment lifecycle — detection, allocation, enforcement, and verification.

      Rating Justification

      • Textual Authority (B): retrieved at abstract level; preprint, not peer-reviewed.
      • Corroboration (4): nineteen authors across academic, civil-society and industry institutions, and — decisively — the levers it describes were subsequently adopted as actual policy instruments (export controls, compute thresholds in regulatory frameworks). Convergence between a paper's proposal and enacted policy is corroboration of the analysis, not merely of the authors' agreement.

      What It Does Not Show

      It does not show compute governance is sufficient, and it does not resolve the tension the vault already carries: INS-007 holds that weights are already in the wild, while compute governance presupposes that the binding chokepoint is training, upstream of distribution. If INS-007 is right, compute governance regulates the production of a good that has already been copied.

      Relation to the Vault

      This is the strongest external rival to SYS-003's chosen chokepoint. The Citadel Stack locates leverage in custody of the inference node; the compute-governance literature locates it in control of the training cluster. These are different bets about where the irreplaceable input sits, and the vault has never stated the comparison. See CNF-007.

      It also supplies the vocabulary the proposal lacks. "Sovereign inference" is not a term of art; compute governance, compute thresholds and export control are. The naming problem recorded in CNF-007 is a symptom of the proposal not being indexed against this literature.

      Back to Shackled AIForests