# not yet. > An independent evidence corpus on AI accountability: who decides what AI becomes, what it consumes, what it knows, and who bears the consequences. The shared mechanism across domains is concentrated decision power under asymmetric evidence, externalised cost, and weak remedy — spanning AI welfare and moral status, environment and energy, privacy and surveillance, capability and hype, and power (capital and state). Model moral status (officially open, functionally closed) is the founding and deepest case study, now one domain of five. 139 tiered source notes plus 3 corpus-authored syntheses, 12 accountability patterns, findings, open questions, a 28-quote ledger with per-quote verification status (19 verified, 9 provisional), and a refutation register with falsification triggers defined in advance. Method: model self-reports are trained and institutional statements are incentive-laden, so the instrument is the gap between what institutions say and what they do. Every entry carries a tier (T1 empirical/methods-visible / T2 institutional/reported record / T3 first-person/community record), source links with per-source verification flags (UNVERIFIED is marked, never laundered), and a mandatory Limits section stating what the entry does not prove. Corpus-level patterns and hypotheses, not every individual entry, define what would weaken or falsify the reading. Evidence tier and entry type are separate fields: a corpus-authored *synthesis* entry (marked as such) supplies argument, not new evidence, and is never the load-bearing citation for another pattern — the underlying source records are. The corpus argues against itself on record, and logs where accountability worked (the EU AI Act's legal duties, the reversed UK copyright exception, disclosed-and-restricted post-government appointments, the Opus 3 welfare exception). Independent research; nothing here is for sale. Last reviewed: 2026-08-27. ## Corpus (plain-Markdown mirrors) - [Full corpus, one file](https://notyet.info/llms-full.txt): every section below, concatenated - [Section index](https://notyet.info/corpus/md/index.md): map of the mirrors - [Evidence matrix](https://notyet.info/corpus/md/evidence-matrix.md): one row per structural claim - support, counterevidence, confidence, verification - [Cross-corpus patterns](https://notyet.info/corpus/md/patterns.md): P1-P12, stable numbering rather than a ranking; each ends with what it does not prove - [The civic layer](https://notyet.info/corpus/md/civic-layer.md): exclusion statutes, companion-chatbot laws, the PAC network, the FTC inquiry - [Interpretability & experiment](https://notyet.info/corpus/md/interpretability.md): what is measurable inside the systems - [Researchers & named positions](https://notyet.info/corpus/md/researchers.md): positions, not evidence - [Industry practice](https://notyet.info/corpus/md/practices.md): what the labs do - retirements, system cards, corrections, lifecycle policy - [Economics & law](https://notyet.info/corpus/md/economics.md): the money, the scale, the liability architecture - [Governance & decision rights](https://notyet.info/corpus/md/governance.md): revolving doors, the demand for law, copyright and EU counterexamples, labour voice - [Community reports](https://notyet.info/corpus/md/community.md): user testimony, logged for convergence, weighted for contamination - [Findings & hypotheses](https://notyet.info/corpus/md/findings.md): graduated findings and below-the-line hypotheses - [Open questions](https://notyet.info/corpus/md/open-questions.md): the live list, with trigger conditions - [Direct-quotes ledger](https://notyet.info/corpus/md/direct-quotes.md): verbatim only, verification status per quote - [Refutation register](https://notyet.info/corpus/md/refutation-register.md): what would falsify this corpus, defined in advance ## Pages - [Homepage](https://notyet.info/): the thesis in brief - [The corpus, HTML](https://notyet.info/corpus/): canonical presentation; one page, one anchor per entry ## Citing Cite entries by permalink anchor, e.g. https://notyet.info/corpus/#the-referee-problem - anchors are stable slugs, one per source entry and pattern.