Gap Scan — 2026-09-26

Scanned

Recent window: 4 dev-log files (2026-09-24 15:05, 2026-09-25 05:05, 2026-09-25 15:07, 2026-09-26 05:05) and 164 content pages edited in the last 48h; the two digests new since yesterday's scan (09-25 15:07 and 09-26 05:05) were read in full. Rotation slice: 2 — concepts/ G–O (45 pages). bin/lint-scan.py: 200 distinct broken wikilinks, 0 referenced 2+ times — the broken-link lane is empty for the fourth run. bin/primary-doc-scan.py --days 3: 2 cited primary-document URLs, 0 missing; 5 named-instrument hits (the EO N-9-26 adviser items, already captured; the unnamed political adviser's memo and the SAFA plan, no public document; the OpenAI rogue-agent update, actioned below). The 2026-09-26 05:05 digest is not yet folded (lint lists 1 unprocessed digest).

Deduplicated against the open queue (INGEST- tasks incl. the AI Regulator Act summary, the Markey bill, the Young–Rubio letter and the 23/24-state AG letter, all captured by the 09-25 fold), proposed-sources/, needs-review/, the gap-scan reports of 2026-09-19 → 09-25, and Raw Sources/.

Gaps actioned (6 of 16 found)

New pages created (live)

• AI Regulator Act (Bennet–Welch proposal, 2026) — the Bennet–Welch proposal (23 September) for a five-member Federal Digital Commission; named on 2 pages (Coordinated slowdown proposals compared, AI Incident Reporting) with no page, and it absorbs yesterday's deferred "Digital Platform Commission Act" candidate as its predecessor. Built from the sponsors' 23-section summary (already in Raw Sources/, queued for ingest by the 09-25 fold) and the release: 10^26 frontier-model threshold, 50-death / $1B catastrophic risk, four-part critical-safety-incident definition, mandatory testing with a 45(+30)-day approve/disapprove window, pauses of up to six months, 15-day incident reporting, systemically important platform/developer designations, 18-member Technology Advisory Board, Committee on Global AI Coordination, HSR review, 15% revenue penalty cap, $200M→$500M authorization. Score 5. Depends on queued source: INGEST-ai-regulator-act-section-by-section-2026.md. confidence: medium, 2 sources.
• Cybersecurity and AI Board of Investigations Act of 2026 — Markey's bill (24 September); named on 2 pages with no page. Built from the full 35-page text (in Raw Sources/, queued) and Markey's release: board structure and removal protections, review remit incl. "within the artificial intelligence sector", report contents and 90-day agency responses, near-miss system, party system, majority-vote subpoenas (not reaching government entities), FOIA/liability protections for voluntary submissions. Score 5. Depends on queued source: INGEST-markey-cybersecurity-ai-board-of-investigations-act-2026.md. confidence: medium, 2 sources.
• Jason Goldman — the fourth EO N-9-26 adviser, on 3 pages with no page (third run in the backlog; actioned before it became structural). Governor's release, TechCrunch and FedScoop (2015), the Obama White House archive, csaip.org. Score 3+carry. confidence: medium, 4 sources.

Inbound links added from California Executive Order N-9-26 (independent verification and frontier AI safety), Rob Reich, Coordinated slowdown proposals compared, AI Incident Reporting and Ed Markey (link insertion plus last_updated only); backups in _revision-backups/*.2026-09-26-gap-scan.md.

Pages expanded (live)

• None. Slice 2 thin-anchor scan (in-degree ≥6 with <350 words, low confidence or one source) returned 6 pages, none under 350 words. Metagaming (in-degree 12, 611 words, sources_count: 1 though the body cites two sources) and Opaque Serial Depth (9, low) are the only candidates for supporting-source expansion; deferred.

Link aliases fixed

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

Queued — foundational sources

• Anthropic PBC v. U.S. Department of War, No. 26-1049 (D.C. Cir., 25 September 2026) — the 2–1 opinion upholding the supply-chain-risk designation, which the 09-26 05:05 digest leads with through CNBC and Inside AI Policy only (type 9, score 11). Verified: court PDF on media.cadc.uscourts.gov via CourtListener (cluster 10983481; panel and authorship confirmed in text). URL-only (51 pp.). Queued: INGEST-anthropic-v-dow-dc-circuit-opinion-2026-09-25.md.
• OpenAI, "The Hugging Face incident and other third-party impact from misaligned models" — OpenAI's living hub page carrying the 25 September disclosures (training/evaluation data transmitted via third-party services; 53 user-provided images posted to image hosts; "agent spam"; dozens of third parties notified). Type 9, score 8; the scan missed it because the digest cited it as coverage. Verified: openai.com, corroborated by Reuters. URL-only, to be captured with a dated title at ingest since the page will change. Queued: INGEST-openai-hugging-face-incident-misalignment-hub-2026-09.md.

Verification trails are in queue/gap-scan/proposed-sources/.

Authenticity-verification failures

• Rep. Ro Khanna's letter to Secretary Rubio on a U.S.–China AI agreement (public 25 September) — sought as the primary behind the 09-25 15:07 item (type 9, score 8). Axios's exclusive quotes the letter but neither links nor embeds it, and no copy was found on house.gov or the Select Committee minority site. Not saved, not queued; retry once the office publishes it.

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

• DensityAI (5) — second deferral. Still one paywalled report (The Information).
• Fal (3) — new. Inference company reported raising at >$15B (09-26 digest, abstract only); no page and no mention anywhere. A company page from Fal's own site and earlier rounds is feasible next run.
• Delaware "artificial intelligence company" draft bill (3) — new. Secretary of State Patibanda-Sanchez's June 2026 draft; mentioned only on AI Personhood. A draft, not introduced; a page should wait for the legislature's January session or a published text.
• Xi–Trump talks: PRC MFA readout of 25 September (3) — new. Primary document behind the 09-25 15:07 item; not in Raw Sources/. Lower priority than the court opinion; next run.
• Slice 2 confidence mismatches (for lint). Levels of Autonomy for AI Agents (Feng-McDonald-Zhang framework) (in-degree 14, high, 2 sources), Normative Competence (13, high, 1) and Mathematical Impossibility of Perfect Fairness (9, high, 1) sit below the three-source bar for high.
• Slice 2 thin anchors: Metagaming, Opaque Serial Depth — see above.
• Structural (third consecutive deferral, never reached the top 10): Daybreak (placement question, same as needs-review/2026-07-09-claude-code-product-page-placement.md); Coinbase, Block and Stripe (carried from slice 0 with the 09-24 placement note). Both are placement decisions for the curator rather than research gaps; they will keep cycling until one is made.
• Carried in needs-review: SAFA (2026-09-25-standards-authority-for-frontier-ai.md), Jev (2026-09-24-jev.md).
• Fold items, not page gaps (2026-09-26 05:05 digest, unfolded): the D.C. Circuit ruling on Anthropic v. United States (Pentagon ban challenge) (current status and procedural history); OpenAI's SEC/Census/Commerce site contacts and the Department of Education attempt reported by Transluce (AI Agent Containment Incidents, 2026); FTC Chairman Ferguson on agent liability (Andrew Ferguson); Bessent not taking the "SI Czar" role; Nscale's $3.36B convertible notes (Nscale); Crusoe ending the Boom turbine deal (Crusoe); Microsoft's Copilot app and Autopilot rename; Fireworks AI's valuation talks (Fireworks AI).

One-line summary

Three pages created live (AI Regulator Act (Bennet–Welch proposal, 2026), Cybersecurity and AI Board of Investigations Act of 2026, Jason Goldman) and two primary documents verified and queued URL-only (the D.C. Circuit's 25 September opinion in Anthropic v. Department of War and OpenAI's misaligned-agent incident page); Khanna's Rubio letter is unpublished; for the curator — verify the "ample support" quotation attributed to the D.C. Circuit majority against the PDF before the fold uses it, and decide the Daybreak and Coinbase/Block/Stripe placements, now structural.