Gap Scan — 2026-06-16

Scanned

Recent window: 4 dev-log files in the last 48h (2026-06-14-2118, 2026-06-14-2205, 2026-06-15-1505, 2026-06-15-2205 — the Anthropic Mythos/Fable export-control standoff and its diplomatic/legal fallout; a children's-AI litigation/legislation cluster; a consumer class action over Claude subscription limits; the launch of the Sequent alignment nonprofit; and a run of industry/market items) plus ~17 wiki pages edited in the last 48h (the software-engineering-labor and EA/longtermism threads carried from the prior two runs). Rotation slice: 12 — sources/ I–P (93 pages).

The dominant live thread remains the June 12 Commerce export-control directive against Anthropic's Mythos 5 / Fable 5; folding the episode itself is developments-log's job and its primary artifacts (NSPM-11, Amodei's "Policy on the AI Exponential," the Lutnick letter) are already ingested or staged, so no page was created for the episode. But the episode surfaced two durable, previously-missing pages — the statute being invoked (ECRA) and the general governance concept it implicates (sovereign AI) — both created this run. The sources/ I–P slice was already well-built; its budget went to the missing-page gaps, the litigation page, the alias fix, and one queued source.

Gaps actioned (6 of ~12 found)

New pages created (live)

• Sequent (Sequent Research) — recent-thread gap (live, score 5): the nonprofit alignment org launched 10 June 2026 by Geoffrey Irving with researchers from UK AISI's Alignment Team and Timaeus, named across the 06-15 dev-log ("alignment is not on track"); no page existed. Built from the founders' own launch post (Alignment Forum / sequent.org) and Import AI 461. confidence: medium. (Leaves Timaeus and Geoffrey Irving as new dangling targets — candidates for a future run.)
• Export Control Reform Act of 2018 (ECRA) — recent-thread + core-area gap (live, score 5): the statute Commerce invoked against Anthropic's models, named in the 06-15 dev-log as "the 2018 Export Control Reform Act" with no legislation/ page. Core US-regulation area. Built from the ECRA primary-law facts (Pub. L. 115-232; §1758 emerging/foundational technologies; permanent EAR authority) corroborated across Akin Gump, EPRS, and MoFo analyses, plus the June 2026 application from Reuters. confidence: medium.
• Sovereign AI — gap type 4 (concept anchored by an ingested source but never created), in-degree 2: AI Sovereignty's Definitional Dilemma — Stanford HAI (2026) explicitly names it "the planned general-concept page Sovereign AI." Distinct from the existing Sovereign AI (Product Concept) (organizational/on-prem sense). Built from the HAI anchor plus the India "sovereign AI" debate the export-control episode reopened. confidence: medium.
• Kahn v. Anthropic PBC — recent-thread gap (live, score 5): the consumer class action over Claude Max 5x/20x usage limits named in the 06-15 dev-log; no litigation/ page. Docket verified via CourtListener (Kahn v. Anthropic PBC, N.D. Cal. 3:26-cv-05763, filed 2026-06-14). confidence: medium.

Pages expanded (live)

• None this run. The rotation slice (sources/ I–P) and the recent-thread anchor pages were adequately built; the budget went to the four new pages, the alias fix, and the queued source.

Link aliases fixed

• [[sources/ft-mundy-americans-dread-ai-2026]] → [[sources/luce-americans-dread-ai]] (3 instances across sources/verge-lopatto-silicon-valley-normal-people-2026-04.md and entities/marc-andreessen.md). Same work — Edward Luce, "Why Americans dread AI," FT, 12 May 2026 — mis-aliased to a wrong author ("Mundy"); the prose author attribution on the Andreessen page was corrected from "Mundy" to "Luce" to match the canonical source page.

Queued — foundational sources

• "Deep-Research Agents Can Be Poisoned via User-Generated Content" (Zhang, Triedman & Shmatikov, Cornell) — recent-thread gap (live, score 6): named in the 06-15 dev-log; no Raw Sources file, no page, no open task. Verified: arXiv (arxiv.org/abs/2605.24245, arXiv:2605.24245v1, DOI 10.48550/arXiv.2605.24245, submitted 22 May 2026). No raw file saved (large arXiv PDF, fetched fresh at ingest). Queued: INGEST-deep-research-agents-poisoned-ugc-2026-06-16.md; verification record in gap-scan/proposed-sources/.

Authenticity-verification failures

• None. The one queued source (the Cornell arXiv preprint) passed verification against its canonical host (title, three-author list, arXiv ID, DOI, submission timestamp, and a distinctive abstract passage all confirmed). The four new pages were built from supporting web sources cited inline as (Source: <URL>) plus the existing [[sources/ai-sovereignty-definitional-dilemma-hai]]; the Kahn docket was confirmed against CourtListener. No raw sources were saved this run.

Data discrepancy flagged for lint

• Cornell preprint date. The 06-15 dev-log frames "Deep-Research Agents Can Be Poisoned…" as released 15 June 2026; the arXiv submission is 22 May 2026 (press coverage appears to be what dated it to June 15). The source page should use the arXiv submission date on ingest.

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

• entities/timaeus — newly relied on by Sequent (Sequent Research); the singular-learning-theory alignment startup. Score 3.
• entities/geoffrey-irving — newly relied on by Sequent (Sequent Research); AISI chief scientist, ex-DeepMind/OpenAI, debate/RLHF contributor. Score 3.
• claude-code — in-degree 8. Perennial; blocked on a products/ placement schema decision. Needs a schema call from the user. Score 3.
• sources/anthropic-pwc-expanded-partnership (in-degree 6) — supporting-class partnership mis-linked as a sources/ page; fold to inline (Source:). Hand to lint. Score 2.
• concepts/frontier-models — a general definitional page distinct from concepts/open-weight-frontier-models / concepts/frontier-ai-governance. Score 2.
• concepts/automated-alignment-research — recurring across alignment pages and now Sequent (Sequent Research); no dedicated page. Score 2.
• In-degree-2 entity stubs: entities/masahiro-mori, entities/anil-seth, entities/lee-anne-fennell, entities/mark-lemley, entities/eric-horvitz, entities/tom-mitchell. Score 1.
• Concept stubs: concepts/iterative-deployment, concepts/ai-system-cards, concepts/ai-and-relationships. Score 1.

Dashboard rebuild

Attempted python3 bin/build-dashboard.py after the four live page creations. Outcome recorded in the audit-log Notes; if it again failed on the iCloud sync-conflict directories under Wiki/_meta/site/, a needs-review note is filed and the live pages remain correct (only the generated HTML is stale).

One-line summary

Created 4 pages live (Sequent, Export Control Reform Act of 2018, Sovereign AI, Kahn v. Anthropic), fixed 1 link alias (3 instances, with an author correction), and queued 1 verified foundational source (the Cornell deep-research-poisoning preprint); flagged the preprint's date for lint. Two new dangling targets (Timaeus, Geoffrey Irving) carried to the deferred backlog.