AI Policy Wiki
Dashboard
⚲
AI Policy Wiki › Briefings

Gap Scan — 2026-10-24

Daily gap hunt — what the wiki is missing and how each gap was triaged.

Contents

Scanned

Recent window: 4 dev-log files (2026-10-22 15:17 → 2026-10-24 05:11; the two newest, 10-23 15:11 and 10-24 05:11, were not covered by yesterday's scan) plus bin/primary-doc-scan.py --days 3 over 7 digests (precise lane 0, named-instrument lane 10). 140 content pages carry a last_updated in the window, but all trace to the 2026-10-23 approved O885/next-57 rollouts (2,148 + 58 article files), which are style/metadata changes, not live threads, so they were not mined as recent-thread signals. Rotation slice: 2 — concepts/ G–O (47 pages; 2 met the thin-anchor screen). bin/lint-scan.py re-run: 53 distinct broken wikilinks (down from 194 before the rollout), 0 referenced 2+ times, so no broken-link candidates and no alias fixes.

Gaps actioned (10 of 45 found)

New pages created (live)

  • Formal Verification — named in prose on 12 content pages (safety cases, AI as Normal Technology, Brockman's "Defender's Window", the Fermat formalization source, Leiden Declaration, ai-for-science, machine-produced mathematics, overview) with no concept page; live thread via Anthropic and Theorem's 2026-10-22 result proving 239 OSS-Fuzz entry points crash-free (10-23 15:11 digest). Built from Anthropic's research post (read in full), the seL4 SOSP 2009 paper and its 2014 comprehensive-verification paper, Tegmark–Omohundro (arXiv 2309.01933), Dalrymple et al. "Towards Guaranteed Safe AI" (arXiv 2405.06624), ARIA's Safeguarded AI programme page and Oxford's ARIA funding note, plus five existing sources/ pages. confidence: medium, 8 sources. No new dependency queued (Anthropic's post is a supporting research post; its appendix PDF could be ingested later if wanted).

Pages expanded (live)

  • Nondelegation Doctrine — slice-2 thin anchor: in-degree 6, 324 body words, confidence: low, 1 source, no account of the case law. Added the J.W. Hampton (1928) intelligible-principle formulation, the two 1935 invalidations (Panama Refining, Schechter), a ## Recent Supreme Court decisions section on Gundy v. United States (2019; plurality, Alito concurrence, Gorsuch dissent) and FCC v. Consumers' Research (2025; 6–3, intelligible-principle test reaffirmed, private-nondelegation claim rejected). All original facts and the CAP citation preserved. 2 new (Source: URL) cites (supremecourt.gov slip opinions 17-6086 and 24-354). sources_count 1 → 3; confidence low → medium. Backup: _revision-backups/nondelegation-doctrine.2026-10-24-gap-scan.md.
  • None — the referenced-2+ list is empty.

Queued — foundational sources (all URL-only; trails in queue/gap-scan/proposed-sources/)

  • UK AISI, "Investigating alignment failures in AISI's incident" (2026-10-23) — lead item of the 10-24 05:11 digest (Mythos 5 GitHub supply-chain incident). Verified: aisi.gov.uk. Queued: INGEST-aisi-investigating-alignment-failures-incident-2026.md.
  • Anthropic, "An alignment assessment of the UK AISI cybertesting incident" (2026-10-23) — coordinated companion. Verified: anthropic.com; one extraction surfaced an "August 4, 2026" date inside the page, flagged for the ingest. Queued: INGEST-anthropic-alignment-assessment-ukaisi-2026.md.
  • OpenAI, "Securing our frontier model training environment" (2026-10-23) — no-live-internet training policy after the Sept. 20 DNS-resolver incident. Verified: openai.com; corroborated by Bloomberg. Queued: INGEST-openai-securing-frontier-model-training-environment-2026.md.
  • Olmo, Zhang & Tracy, "Simulated Control Evals" (arXiv 2610.31593) — Redwood control-evaluation paper. Verified: arXiv + Redwood blog. Author correction recorded: the digest omitted first author Jeffrey Olmo. Queued: INGEST-olmo-zhang-tracy-simulated-control-evals-2026.md.
  • Treasury COINS Act NPRM, "Outbound Investment Security Program Regulations" (2026-10-22) — the 10-23 15:11 item cited only The Information. Verified: home.treasury.gov PDF (Docket TREAS-DO-2026-0595, RIN 1505-AC94, 163 pp., comments due Dec. 7) and press release sb0667; not yet in the Federal Register. Uses "SI systems" terminology. Queued: INGEST-treasury-coins-act-outbound-nprm-2026.md.
  • Microsoft Threat Intelligence, Storm-0940 / ARTEX / ConnectWise report (2026-10-21) — top-8 deferral from 10-23. Verified: microsoft.com; corroborated by CyberScoop, The Record, Dark Reading. Queued: INGEST-microsoft-storm-0940-artex-connectwise-2026.md.
  • Warner, Padilla, Himes and Morelle letter to 25 technology CEOs (2026-10-19) — top-8 deferral from 10-23. Verified: 7-page PDF on padilla.senate.gov; corroborated by Warner's press release and Democracy Works. Queued: INGEST-warner-padilla-himes-morelle-election-letter-2026-10-19.md.
  • SEC Division of Examinations, FY2027 Examination Priorities (2026-10-23) — 10-24 05:11 item. Verified: sec.gov PDF (20 pp.) and press release 2026-109. Queued: INGEST-sec-fy2027-examination-priorities.md.

Authenticity-verification failures

  • None this run.

Deferred backlog (over the daily cap — re-surfaces next run)

  • Microsoft AI MAI-DxO Science study (6) — new. 10-23 15:11; no page anywhere; the Science paper is the primary. Queue first next run.
  • Kimi K3.1 Pro (5) — new. Fold into Kimi K3 via developments-log, or a models/ page via model-robustness-check; weights due by Nov. 4.
  • Harrigan–Whitesides anti-distillation bill (5) — new. Bill number and text not located; Adversarial Distillation exists.
  • Anthropic draft S-1 (5) — carried in substance. Not publicly filed; fold item, not a page gap.
  • Epoch brief: SpireClimb, "Can AI automate Epoch?", Cyber ECI (4) — new. Single newsletter source; wait for a benchmark page or paper.
  • Brennan Center chatbot election-image follow-up (4) — new. Unenriched in the digest.
  • AAIF's Open Responses, OpenEnv, OpenSandbox (3) — new. Fold into Agentic AI Foundation (AAIF).
  • Hg–Anthropic European consultancy (3), SK Hyper 15 GW plan (3), Applied Intuition Navy contract (2), Sofia Liguori resignation (2) — new. Fold items.
  • Anthropic draft federal frontier-AI bill (5), DoD Technology Support Office (5), Project Agincourt (5), ARCYBER ECHOTRIBBLE (5), New Jersey A5711 / S4809 (5) — carried from 10-23.
  • Guardrails Action coalition letter (6), OpenAI mathematics results / IAS advisory-group recommendations (6), National Compute Grid consortium paper (6), letter from three fired OpenAI employees on CoT monitorability (6) — third consecutive deferral (10-08, 10-23, 10-24). These are now structural: each scores 6 but is outranked by a steady supply of fresh 8s. Reported once here; consider a backlog sweep or a deliberate slot next run.
  • Rep. Trahan's CLAIM Act (5), Kyoto Vision (5); Reflection AI Beam, Scott Kupor, Internet Watch Foundation, Biohub / Virtual Biology Initiative, personal agent protocol, Conrad Stosz, Delaware AI-company draft bill (3 each) — carried.
  • Thin anchors: LLM fallacy (slice 2; 311 words, in-degree 8, medium on 2 sources); carried from slice 1: AI as Cultural Technology, AI Policy, China’s generative AI filing and registration system.
  • Structural (seventh consecutive deferral, reported once): Daybreak placement; Coinbase, Block and Stripe placement — curator decisions, not research gaps.

Observations for the curator

  • .git/index.lock still present (zero bytes, 2026-10-05 15:00 UTC; 19 days). The 10-23 rollout log says it left "the checkout's preexisting sync conflict, original HEAD/index and active Git process" intact, so the lock may belong to a live process — check before rm -f .git/index.lock. This run held the vault lock in file-only mode and ran no git commands.
  • Developments-log fold still stalled: lint now counts 48 unprocessed digests (2026-09-28 → 2026-10-24). Most of the deferred backlog above is fold work.
  • Rollout interaction: neither page touched today was in the 10-23 rollout sets (the nondelegation page's prior last_updated was 2026-09-26; the formal-verification page is new), so no rollout hold should be affected.

One-line summary

One concept page created (Formal Verification), one thin anchor expanded (Nondelegation Doctrine), eight primary sources verified and queued URL-only (AISI and Anthropic incident analyses, OpenAI training containment, Redwood SCE paper, Treasury COINS NPRM, Microsoft Storm-0940, Warner et al. election letter, SEC FY2027 priorities); for the curator — the four 6-score items are now structurally stuck, and the git lock and 48-digest fold backlog remain.

Categories: Briefings | briefing | gap-scan