AI Policy Wiki
Dashboard
⚲
AI Policy Wiki › Briefings

Gap Scan — 2026-07-12

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

Contents

Scanned

Recent window: 4 dev-log files (07-10 morning/evening, 07-11 morning/evening), ~47 pages edited in last 48h. Rotation slice: 10 — analysis/ (22 pages). Note: the 07-11 evening dev-log (22:06) remains unprocessed for tonight's developments-log cycle; only its gap-relevant items were actioned here (07-10/07-11 precedent) — its non-gap folds (Grok 4.5, China Claude Code alert, FT Singapore model-access story, Karp quotes, GPT-5.6 tier pricing, etc.) were left for the nightly cycle.

Gaps actioned (7 of ~18 found)

New pages created (live)

  • Tencent / Hunyuan — type-3 missing company page on a live thread (score ~5: +3 live Hy3 open-source release in the 07-11 evening dev-log, +2 US-China core area). Tencent was named across 16 wiki pages (Manus unwind investor, $20B DeepSeek-centric infrastructure with Alibaba, H200 clearance list, Pentagon blacklist, Kling AI backer, Kunlunxin customer) with no page anywhere. Built from Tencent's own release announcement (primary host), Simon Willison, VentureBeat, and TechTimes: Hy3 (Hunyuan 3.0) July 6, 2026 — 295B/21B-active MoE + 3.8B MTP, 256K context, Apache 2.0 dropping the EU/UK-blocking license restrictions, vendor benchmark claims vs GLM-5.2 / DeepSeek-V4 / GPT-5.5. Backlinked at five existing mentions (manus-butterfly-effect, circular-financing-ai, kunlunxin, kuaishou-kling, us-china-ai-competition) and added to Index. confidence: medium.

Pages expanded (live)

  • AI Copyright Litigation — Analysis — top slice thin anchor: in-degree 58 on medium/4-sources, stale since 06-06 while the docket moved. Added the May 14, 2026 Bartz fairness hearing (Judge Martínez-Olguín; 92.77% claims rate; $3,000–$3,100/work; supplemental brief ordered on late opt-outs; approval pending as of June per Clark Hill), the 28-author opt-out suit (May 14, jury-trial demand), and the NYT 2026 procedural record folded from NYT v. Microsoft, OpenAI et al. (Jan 5 20-million-log order; SJ briefing closed April 2; July 9 sanctions motion). sources_count 4 → 11. [7 new (Source: URL) cites]
  • Legal Services — AI Deployment — deferred carry (07-11) turned live: in-degree 19, 673 words, stale 06-06, and the 07-11 evening dev-log's UChicago item touched it directly. Added a hallucinated-citations section (1,000+ flagged filings vs ~65M annual cases) and a legal-education section (UChicago Law AI strategy with 1L device ban, primary host; Sinofsky's Harvard-1982 analogy, attributed). sources_count 4 → 6. [2 new (Source: URL) cites]
  • DeepSeek-R1 — deferred carry: 539 words at in-degree 41, stale 06-06, missing parameters/safety_case and a Relationships section. Added a Publication and peer review section (Nature cover, Sept 18, 2025, vol. 645 — described in coverage as the first mainstream LLM through journal peer review; $294K disclosed RL training cost), a DeepSeek V4 Pro / V4 Flash lineage link, typed Relationships; frontmatter repaired. sources_count 3 → 5. [3 new (Source: URL) cites]
  • Kimi K2 — deferred carry and the high-on-sources_count-2 pattern (in-degree 30). Found a genuine source conflict on the weight license: the arXiv technical report (and the wiki's source page) say CC BY-NC-ND 4.0; the Hugging Face release and contemporaneous coverage (Willison, Interconnects, VentureBeat) say Modified MIT. Licensing section now presents both, attributed. Added Kimi K2 Thinking (Nov 6, 2025, Modified MIT) to Related models; frontmatter repaired (parameters, safety_case; sources_count 2 → 5). Confidence lowered high → medium pending resolution of the license conflict — flagged for lint/ingest attention. [3 new (Source: URL) cites]
  • Gemini 3 / Gemini 3 Pro — deferred carry: 725 words for Google's flagship family at in-degree 26, stale 06-06, coverage ended at an undated "3.1 Pro refresh". Added the 3.1 Pro release date (Feb 19, 2026, Google's own post), the April Deep Research agents, and a Gemini 3.5 series section (I/O May 19, 2026: 3.5 Flash benchmarks, default-model status, Gemini Spark, Frontier Safety Framework statement; 3.5 Pro reported targeting July 17). sources_count 5 → 9. [4 new (Source: URL) cites]
  • None this run — the top broken-link targets are either queued ingests (sources/trustworthy-agents-in-practice, sources/claude-sonnet-5-system-card), curator-blocked (claude-code), or standing low-priority carries (concepts/foundation-models, person pages at in-degree 2).

Queued — foundational sources

  • Palantir, Institutional Sovereignty in the Age of AI (2026) — type-1 dangling foundational reference on a live thread (WSJ 07-11 feature; named by title, advances an original 15-step framework and the "extraction-prone models" term). Verified: canonical PDF on palantir.com's own asset host; distinctive passages confirmed by direct PDF parse; corroborated by WSJ, Palantir's X post, and an SCSP briefing mention. Queued URL-only (28-page PDF): INGEST-palantir-institutional-sovereignty-2026-07-12.md (+ verification record in proposed-sources/). A supporting mention with a provenance note was folded into Palantir Technologies (sources 8 → 10).

Authenticity-verification failures

  • None. The one pulled source verified on its primary host at the first attempt.

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

  • Bartz v. Anthropic — stale (06-06); the May 14 fairness-hearing record folded into the analysis page today belongs on the case page too. Score ~4.
  • US AI Regulatory Approaches Compared (in-deg 36, medium, stale 06-06, missing sources_count) and Labor Disruption Timelines: Who Predicts What and Why (in-deg 21, same pattern) — next slice-10 thin anchors. Score ~3–4 each.
  • Sora and Veo (Video Generation Models) (high on src 2, in-deg 21) — the kimi-k2/qwen3 frontmatter pattern, carried again. Score ~4.
  • Possible standalone models/gemini-3-5 page once 3.5 Pro ships (reported July 17) — folded into the family page for now. Score ~3.
  • Anti-AI protest movement / Stop AI (WSJ 07-11 feature; Quinnipiac 70%/55% poll) — zero wiki coverage; single touch, page if references recur (Markey precedent). Score ~3.
  • ChinaTalk "Eastern Data, Western Compute" CAICT analysis — carried from 07-11. Score ~3.
  • Helix (Figure AI Vision-Language-Action model) (257w, medium/1, in-deg 6) — carried; confirm substance next models slice. Score ~3.
  • Markey "AI Accountability Agenda" primary document and People-First Chatbot Act bill text (once on congress.gov) — lint-suggested source queue candidates. Score ~2–3.
  • Frontmatter hygiene: six analysis/ pages lack sources_count (us-china-ai-competition, us-regulatory-approaches-compared, labor-disruption-timelines, safety-frameworks-compared, eu-vs-us-ai-regulation, interpretability-and-safety) — lint territory, noted for the next lint run.
  • Standing carries: claude-code/claude-cowork placement (curator-blocked), inverse-cooking/inverse-trust coined terms (needs-review), duplicate entities/cdao/government/cdao, concepts/foundation-models disambiguation, person/org pages at in-degree 2. Score ~2–3.

One-line summary

Tencent finally has a company page anchored on the Hy3 Apache-2.0 release, the wiki's most-linked stale analysis page (ai-copyright-litigation, in-degree 58) now carries the Bartz fairness hearing and the NYT sanctions motion, four more deferred thin anchors (industries/legal, deepseek-r1, kimi-k2, gemini-3) were strengthened with a genuine Kimi K2 license conflict surfaced and flagged, and Palantir's Institutional Sovereignty white paper was verified on its primary host and queued — user review needed on the queued ingest and the K2 license conflict.

Categories: Briefings | briefing | gap-scan