AI Policy Wiki
Dashboard
⚲
AI Policy Wiki › Briefings

Gap Scan — 2026-09-25

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

Contents

Scanned

Recent window: 4 dev-log files (2026-09-23 15:06, 2026-09-24 05:06, 2026-09-24 15:05, 2026-09-25 05:05) and the pages edited in the last 48h; the two digests new since yesterday's scan were read in full. Rotation slice: 1 — concepts/ A–F (196 pages). bin/lint-scan.py: 206 distinct broken wikilinks, 0 referenced 2+ times — the broken-link lane is empty for the third run. bin/primary-doc-scan.py --days 3: 1 cited primary-document URL missing (Liccardo outline — actioned, lane now at zero); 4 named-instrument hits, of which the White House political adviser's EA memo and the SAFA plan have no public document, Newsom's EO N-9-26 is already captured, and the CFTC letter to CME was not published. The 2026-09-25 05:05 digest is not yet folded by developments-log (lint lists it as unprocessed).

Deduplicated against the open queue (113 INGEST- tasks), proposed-sources/, needs-review/, the gap-scan reports of 2026-09-18 → 09-24, and Raw Sources/. Dropped as already in flight: the Warner–Coons–Reed letter to Hegseth and Clayton, Van Nieuwerburgh's Brookings paper, and Transluce's urlquery study (all queued); Basecamp Research (page created by the nightly fold since yesterday).

Gaps actioned (7 of 16 found)

New pages created (live)

  • Pennsylvania HB 2800 (Artificial Intelligence Risk Prevention Act) — the Artificial Intelligence Risk Prevention Act, introduced 22 September 2026 by Rep. Melissa Shusterman with 14 Democratic co-sponsors and referred on 23 September; live in the 05:05 digest, no page or mention anywhere. Built from the printed bill (PN 3904, 27 pp.), the General Assembly bill page and the sponsors' 14 September release: definitions (10^26 operations, $500M revenue, the 50-death / $1B catastrophic-risk threshold, four-part critical-safety-incident definition), registration and beneficial-ownership disclosure from 1 January 2027, frameworks and annual third-party audits from 1 January 2028, 72-hour / 24-hour incident reporting, whistleblower protections, PEMA/AG enforcement with $1M / $3M per-violation ceilings and $1,000 per day for unregistered operation, no private right of action. Score 5. Depends on queued source: the bill text. confidence: medium, 3 sources, ~2,190 words. Inbound link added from State-Level AI Regulation (frontier-lab transparency row).
  • EU Product Liability Directive (Directive (EU) 2024/2853) — Directive (EU) 2024/2853, named on 8 content pages with no page (deferred 2026-09-24, re-surfaced per step 3); applies from 9 December 2026 and is the enacted half of the 2022 package whose other half, the AI Liability Directive, got its page yesterday. Covers timeline, software/AI-as-product (recital 13), damage, defectiveness incl. continuous learning, liable operators, disclosure and presumptions, the development-risk defence and its Article 18 derogation, limitation, and the relation to the withdrawn AILD. Score 5. Depends on queued source: the directive text. confidence: medium, 4 sources, ~1,700 words. Several article-level statements rest on the EUR-Lex summary and a Latham & Watkins alert; the ingest task says which. Inbound links added from EU AI Liability Directive (proposed, withdrawn) and Specification Hazard.
  • Akamai Technologies — Anthropic's $11.6B / seven-year compute commitment and warrant for up to ~5% of Akamai (announced 24 September) made the company live; named on 6 pages with no page. Built from the Form 8-K and its Exhibit 99.1 on SEC EDGAR, Akamai's history and leadership pages, the October 2025 Inference Cloud release, Bloomberg (May deal) and Investing.com. company_type: cloud-infra. Score 5. confidence: medium, 7 sources, ~1,100 words. Inbound links added from Anthropic (Compute snapshot row) and AI Data Centers.
    • Fidelity note for the nightly fold: the 05:05 digest says Akamai "signed" the agreement on 24 September. The 8-K dates Project Plans 2 and 3 and the warrant to 18 September; 24 September is the announcement. The page records both. Anthropic's Compute snapshot still lacks the September row — fold work, not added here.
  • Rob Reich — one of Newsom's four EO N-9-26 advisers (23 September); on 3 pages with no page, deferred 2026-09-24. Stanford political-science profile plus the governor's announcement. Score 3 (raised by the carried deferral and the live EO thread). confidence: medium, 3 sources. Inbound link added from California Executive Order N-9-26 (independent verification and frontier AI safety).

Backups of every edited existing page are in _meta/_revision-backups/*.2026-09-25-gap-scan.md; each edit is a link insertion plus last_updated.

Pages expanded (live)

  • None. The referenced-2+ list is empty.

Queued — foundational sources

  • "A Call for Control of Frontier AI Models" (Støre, Stubb et al., 21 September 2026) — the joint heads-of-government statement behind the 05:05 item, which cited no primary URL (type 9, score 8). 29 endorsers in the captured version, including Albanese, Carney, Macron, Merz and von der Leyen; the US, UK and China are not among them. Verified: regjeringen.no, corroborated by government.nl and the ANI report of Guterres's welcome. Saved: Raw Sources/A Call for Control of Frontier AI Models (Store, Stubb et al., 2026-09-21).md (full text). Queued: INGEST-call-for-control-frontier-ai-models-2026.md.
  • Liccardo, "Negotiating AI Safety Globally" bill outline — the first-lane primary-doc-scan.py hit (type 9, score 8). Verified: liccardo.house.gov, corroborated by the office's press release and an IANS report of 23 September. Saved: Raw Sources/Negotiating AI Safety Globally - International AI cooperation bill outline (Liccardo, 2026-09).md (full text; text layer reads "Al" for "AI", recorded in capture_note). Queued: INGEST-liccardo-international-ai-cooperation-outline-2026.md. Fidelity note: the digest attributes the bill to Liccardo alone and says no text is out; the press release calls it bipartisan with Rep. Kevin Kiley (I-Calif.) — the ingest should find the bill number on congress.gov.
  • Pennsylvania HB 2800, PN 3904 — backs the new legislation page. Verified: palegis.us (the older legis.state.pa.us text URL 404s). URL-only. Queued: INGEST-pennsylvania-hb-2800-text-2026.md.
  • Directive (EU) 2024/2853 — backs the new PLD page. Verified: EUR-Lex (ELI, CELEX 32024L2853). URL-only. Queued: INGEST-eu-product-liability-directive-2024-2853.md.

The two short documents were saved in full; the two long ones are URL-only, since re-typing a 27-page bill or an Official Journal directive from tool output would risk the fidelity the raw record guarantees. Verification trails are in queue/gap-scan/proposed-sources/.

Deferred under the quality gate

  • Standards Authority for Frontier AI (SAFA) — the Google–OpenAI–Anthropic self-regulatory body (score 5; 4 pages). Every account traces to one paywalled report in The Information; the name is tentative, leadership is under discussion, and nothing has been chartered. Note left at needs-review/2026-09-25-standards-authority-for-frontier-ai.md with the conditions for creating the page.

Authenticity-verification failures

  • None this run. The Trump UNGA-81 text remains the standing structural item from 2026-09-24 (no official full text; curator decision pending) and was not retried.

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

  • DensityAI (5) — new. Former Tesla Dojo leaders' chip startup, reported in talks at a $10B valuation with an AWS purchase agreement; no page. One report (The Information); the fold should place it or the next scan should re-test.
  • Digital Platform Commission Act (5) — new. Bennet and Welch reintroducing it (23 September); one mention (Coordinated slowdown proposals compared). The Inside AI Policy item is lede-only and no reintroduced text was found; the prior-Congress text on congress.gov would be the base for a page.
  • Jason Goldman (3) — second deferral. The fourth EO N-9-26 adviser, still without a page; a quick create next run now that Reich's page exists.
  • Daybreak (3) — second deferral, placement question (same as needs-review/2026-07-09-claude-code-product-page-placement.md).
  • Slice 1 confidence mismatches (2) — new, for lint. AI as Social Technology (in-degree 18), AI Policy (umbrella) (9) and Anticipatory AI Ethics (Lazar framework, including the Technological Horizon) (8) carry confidence: high on sources_count: 1, below the three-source rule. AI Consciousness (in-degree 13, low, 554 words) is the one slice-1 page where an expansion from supporting sources would be legitimate.
  • Coinbase (1), Block (1), Stripe (1) — second deferral, carried from slice 0 with yesterday's placement note.
  • Carried: Jev (needs-review/2026-09-24-jev.md).
  • Fold items, not page gaps (2026-09-25 05:05 digest, unfolded): Anthropic's founder-control share class; the Akamai row on Anthropic; Oracle's force-majeure notice to STACK on Project Jupiter (Oracle); NSA's AI Security Center spending (National Security Agency (NSA)); the White House request that OpenAI and Anthropic hold new models from UK AISI pending US testing; Trump–Johnson CEO meeting of 29 September; Kratsios at the UN Security Council.

One-line summary

Four pages created live (Pennsylvania HB 2800 (Artificial Intelligence Risk Prevention Act), EU Product Liability Directive (Directive (EU) 2024/2853), Akamai Technologies, Rob Reich) and four foundational sources verified and queued (two saved in full: the Støre–Stubb "Call for Control of Frontier AI Models" and Liccardo's U.S.–China AI bill outline; two URL-only: PA HB 2800 and the PLD); SAFA deferred to needs-review; for the curator — the Liccardo bill is bipartisan with Kiley, which the digest missed, and the Akamai deal was entered 18 September, not 24.

Categories: Briefings | briefing | gap-scan