Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).
This registration establishes a timestamped, immutable record of a working paper analyzing the structural conflict of interest that arises when frontier AI providers simultaneously serve as scientific infrastructure and as competitors in the research frontier their infrastructure enables. Using the September 2026 Navier–Stokes episode as a stylized case (taking no position on its disputed facts), the paper argues that the resulting problem is one of verifiability, not virtue: when a provider controls the model, the logs, the data flows, and a competing research program, its assurance that it did not use a researcher's work cannot be independently checked. The paper specifies: a six-risk threat model; five design principles; a four-rung "credibility ladder" of institutional arrangements; a full specification of the recommended baseline (a legally separate, externally governed research-service entity with legal, personnel, infrastructure, data, and audit separation); evidence mechanisms including append-only transparency logs and priority timestamps; peer-reviewed allocation of frontier inference; and a pre-committed escalation clause converting provider control into third-party custody upon repeated breach. Precedents are drawn from telecommunications (BT–Openreach), energy regulation, securities law (the 2003 Global Research Analyst Settlement), data governance, and platform regulation (the EU Digital Markets Act). A condensed version of this argument, written for a general scientific readership, is available as a companion policy essay (currently under editorial review) and is registered separately. Tags： AI governance, scientific infrastructure, research integrity, structural separation, auditability, compute allocation, priority in science, frontier AI
No comments yet — start the discussion below.