AI Policy Wiki
Dashboard
⚲
AI Policy Wiki › Briefings

Gap Scan — 2026-08-20

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

Contents

Scanned

Recent window: 4 New Developments Log/ files from the last 48h (2026-08-18 08:05 and 22:05; 2026-08-19 08:05 and 22:08). Rotation slice: 7 — legislation/ (all), 141 pages.

Broken-link analysis taken from bin/lint-scan.py output rather than a hand-rolled regex, per the skill: 243 distinct broken targets across 268 refs at the start of the run, 25 referenced 2+ times, 218 referenced once. Candidate generation was restricted to the referenced-2+ list plus dangling-reference and recent-thread scans; the single-reference long tail was not re-audited.

The slice and the deferred backlog converged this run. The 2026-08-17 report identified six genuinely missing US-law anchor pages and called them "the natural next batch"; four of the six are legislation or litigation pages, and today's rotation slice is legislation/. That batch was actioned rather than deferred a third time.

Gaps actioned (7 of 21 found)

New pages created (live)

  • Children's Online Privacy Protection Act (COPPA) — score 5 (in-degree ≥10 +2, wiki core area +2, lint/deferred backlog +1). "COPPA" appears across 17 content pages including Kids Online Safety Act (KOSA, S. 1748), Kids Internet and Digital Safety Act (KIDS Act, H.R. 7757), Algorithmic Speech Doctrine, Three Theories of Consent Failure (Information / Capacity / Design) and eight Farahany class sources, with no page and no alias anywhere in the vault. Built from the FTC's COPPA Rule page and the April 22, 2025 Federal Register final rule (RIN 3084-AB20, effective June 23, 2025, full compliance April 22, 2026): the statutory structure and actual-knowledge standard, the full rulemaking history, and each substantive 2025 amendment — biometric identifiers and government identifiers added to "personal information", the new "mixed audience" definition and its neutral-age-collection requirement, separate verifiable parental consent for third-party disclosure under § 312.5(a)(2), the § 312.10 prohibition on indefinite retention and written retention policy, the § 312.8 written security program, safe-harbor reporting changes at § 312.11(d)(1), and the three newly codified consent methods. Includes the February 25, 2026 FTC age-verification policy statement and a section on how the amended rule reaches voice interfaces, training on user interactions, and third-party inference providers. 1,330 words, confidence: medium, 4 sources.
  • Minnesota HF 1606 (Nudification Technology Prohibition) — score 5 (live thread +3, US AI regulation core +2). "HF1606" had zero occurrences anywhere in the content folders despite the statute being the subject of the xAI First Amendment challenge and the August 17, 2026 Justice Department statement of interest carried in the 2026-08-19 22:08 digest. Alias resolution confirmed it is not covered by State Deepfake Statutes (MN, WA, TX, CA), which treats Minnesota's election-deepfake statute (Minn. Stat. § 609.771) — a criminal prohibition on disseminating a deepfake — while HF 1606 is a commerce-chapter prohibition on the nudification tool, reaching the service operator rather than the publisher. Built from the full second-engrossment text on the Revisor of Statutes site: all seven subdivisions, the "technical skill" exemption, the treble-damages civil action, the attorney general penalty and its Office of Justice Programs appropriation, the venue and section 230 construction clauses, and the August 1, 2026 effective date; plus the xAI complaint language and the DOJ filing. 1,145 words, confidence: medium, 4 sources. Depends on queued source: Minnesota HF 1606, 2nd Engrossment.
  • Moody v. NetChoice, LLC — score 5 (in-degree 10 +2, wiki core area +2, deferred backlog +1). Named on ten content pages, including a case table on Algorithmic Speech Doctrine built around it, with no page. Note the wiki already carries NetChoice v. Bonta (CAADCA litigation), a different case; the distinction is stated on the new page. Built from the slip opinion: both statutes' operative provisions, the vacate-and-remand disposition and its reasoning, the editorial-control holding quoted verbatim, the Zauderer framing of the individualized-explanation provisions with both circuits' opposing findings, and the four separate writings. 914 words, confidence: medium, 2 sources. Depends on queued source: the slip opinion.
  • Authors Guild v. OpenAI — score 4 (in-degree 9 +1, wiki core area +2, deferred backlog +1). Named on nine pages including AI Copyright Litigation — Analysis, AI Copyright and Anderson v. Microsoft — Shareholder Derivative Complaint (June 2026) — which calls it "the first major class action against Microsoft" of its kind — with no case-tracking page. Built from the Authors Guild filing announcement, the docket tracker, and Judge Stein's October 27, 2025 order: the three infringement theories, the consolidation and MDL 3143 centralization history, the Basbanes stay, and the output-infringement ruling including the incorporation-by-reference and stipulation questions, the "more discerning observer" test, and the distinction from the news-summary decision. 911 words, confidence: medium, 4 sources.
  • Jakub Pachocki — score 7, the highest of the run (live thread +3, in-degree 8 +1, wiki core area +2, deferred backlog +1). Deferred on 2026-08-19 as entities/jakob-pachocki. The deferred slug was misspelled. Press coverage renders the name "Jakob"; OpenAI, his published papers and the wiki's own existing sources/ pages all use Jakub. The page was created at the correct slug with a note on the variant, which is why the earlier candidate never resolved against existing prose. Built from the Wikipedia biography for background and competitive-programming record, plus the August 18, 2026 Axios briefing report. 559 words, confidence: medium, 4 sources.

Pages expanded (live)

None. The run's budget went to the seven new pages and their graph wiring; the slice's thin anchors are listed in the deferred backlog below.

Graph wiring

Fourteen inbound links added across eleven existing pages so the new pages are not orphans and the named-but-unlinked references now resolve: AI Liability, Amazon v. Perplexity AI, Training Data Walls and Pay-per-Crawl (Pigouvian Pricing of Agent Traffic) → legislation/cfaa; Kids Online Safety Act (KOSA, S. 1748), Kids Internet and Digital Safety Act (KIDS Act, H.R. 7757) and Algorithmic Speech Doctrine → legislation/coppa; Algorithmic Speech Doctrine → litigation/moody-v-netchoice; AI Copyright and AI Copyright Litigation — Analysis → litigation/authors-guild-v-openai; Chip Smuggling and Export-Control Evasion and Three Theories of Victory (US / EU / China AI Governance) → concepts/foreign-direct-product-rule; OpenAI → entities/jakub-pachocki; State Deepfake Statutes (MN, WA, TX, CA) → legislation/minnesota-hf1606, with a sentence distinguishing the two Minnesota regimes.

Zero, as expected. The settled finding recorded since 2026-08-14 holds: the remaining referenced-2+ broken targets are author surnames and institution shorthand, not aliases of existing pages. The list was not re-audited.

Queued — foundational sources

  • Minnesota HF 1606, 2nd Engrossment (Minnesota Office of the Revisor of Statutes) — primary text for the statute now under First Amendment challenge, named nowhere in the wiki before this run. Verified: revisor.mn.gov, the official state publisher; live fetch, full text, version list consistent with the bill history, distinctive passages confirmed, corroborated by CBS News Minnesota and MPR News coverage of the xAI suit. Saved: Raw Sources/Minnesota HF 1606 - Prohibition on Nudification Technology (2nd Engrossment, 94th Legislature).md. Queued: INGEST-minnesota-hf1606.md. Verification record: queue/gap-scan/proposed-sources/minnesota-hf1606-2nd-engrossment.md.
  • Moody v. NetChoice, LLC, slip opinion, 603 U.S. 707 (2024) — the controlling authority behind the editorial-control argument used against every AI output-regulation statute the wiki tracks; named on ten pages, never ingested. Verified: supremecourt.gov, the Court's own slip opinion; caption, companion case, docket numbers, decision date, opinion author and all four separate writings confirmed; distinctive passages confirmed. Queued URL-only for a fresh fetch at ingest (96-page PDF), per the QUEUE rule. Queued: INGEST-moody-v-netchoice-opinion.md. Verification record: queue/gap-scan/proposed-sources/moody-v-netchoice-slip-opinion.md.

Authenticity-verification failures

None. Two sources were pulled from JS-rendered hosts that returned page shells rather than content — uscode.house.gov for 18 U.S.C. § 1030, and the Cornell LII copy at full length. The Cornell page was re-fetched in structured-query mode and returned the statutory text, which was used for the § 1030 definitions and offense table; that is a secondary host for a primary text, and the page cites it as such rather than claiming the OLRC copy. No source was rejected.

Source-fidelity notes

Three compression errors in secondary coverage were caught against primary text and not carried into the wiki:

  • Penalty unit. Press coverage and the 2026-08-19 digest describe Minnesota HF 1606 as imposing "penalties of up to $500,000 per violation." The statute says "for each unlawful access, download, or use under subdivision 2" — a different and narrower unit of account. The statutory formulation is used on the page and flagged in the ingest task.
  • Name spelling. The deferred candidate entities/jakob-pachocki carried the press spelling. The subject's name is Jakub Pachocki, as used by OpenAI, in his papers, and in five existing sources/ pages. Created at the correct slug.
  • Disposition. Algorithmic Speech Doctrine's case table describes Moody as a "Kagan plurality." The slip opinion is styled the opinion of the Court, delivered by Justice Kagan, on the parts relevant to that claim. The characterization was not corrected on that page this run — correcting an existing page's doctrinal claim is a revision requiring the full fidelity checklist, not a wiring edit — but it is recorded in INGEST-moody-v-netchoice-opinion.md as a check for the ingest.

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

Third consecutive deferral, now structural rather than transient. Two items from the 2026-08-17 "natural next batch" were not reached again: concepts/p-doom (score 3; zero coverage anywhere in the wiki, and now the only member of that batch never actioned) and entities/johns-hopkins (score 3; "Johns Hopkins" appears on 23 pages, entities/sais is unresolvable shorthand for it, and the 2026-08-19 report noted the page appears in an earlier scan's actioned list but does not exist on disk — that discrepancy should be checked before the page is rebuilt). Both are cheap; neither has ever outscored a live-thread item. They are named here once rather than re-listed silently.

Slice-7 thin anchors, none reached (score 3–4 each). legislation/science-new-golden-age (in-degree 6, 208 words, medium/4 — the thinnest page in the slice), legislation/genesis-mission (8, 300 words), legislation/people-first-chatbot-act (8, 330 words), legislation/scale-act-moolenaar (11, 523 words, medium/1 — the highest in-degree of the group and the best single expand candidate), legislation/lieu-obernolte-ai-bill (8, 455 words, medium/1), legislation/trump-dpa-determination-2026-10 (8, 542 words, medium/1). Six pages; the slice is otherwise in good shape.

A sources_count artifact, not a gap. Seventeen legislation pages carry sources_count: 1 with confidence: high and 600–1,500 words, all stamped last_updated: 2026-06-06 — including BIS Framework for AI Diffusion — Interim Final Rule (RESCINDED) (in-degree 38), Paris AI Action Summit Declaration (2025) (35), China — Interim Measures for the Management of Generative AI Services (35), China — Provisions on the Administration of Deep Synthesis Internet Information Services (33), G7 Hiroshima Code of Conduct for Advanced AI (2023) (31), California SB 1047 (VETOED) (25) and Texas Responsible AI Governance Act (TRAIGA / HB 149) (25). These are well-built pages whose frontmatter under-counts their sourcing, a residue of the 2026-06-06 batch. They read as thin anchors to any in-degree-plus-sources_count filter and will keep surfacing until the counts are repaired. That is a lint task, not a gap-identifier one, and is recorded here for routing rather than actioned.

Recent-thread items with no page (score 3–4). government/house-office-of-legislative-counsel (carried from 2026-08-19), Private Safety Processing (OpenAI's zero-data-retention safety signal, announced 2026-08-19, with a technical white paper promised for September 2026 — likely folds into an existing privacy or safety concept page rather than warranting its own), the Stripe–OpenRouter acquisition and Stripe's "beginning of the singularity" framing (both Stripe and OpenRouter exist; this is a developments-log fold, not a gap), the Irregular postmortem criticism (page exists at Irregular; also a fold), and Senator Adam Schiff's AI-taxation position (page exists at Adam Schiff; a fold).

Broken-link 2+ candidates not reached (score 2–3 each), unchanged. entities/kevin-klyman, entities/michael-j-d-vermeer, entities/lee-anne-fennell, entities/sacha-altay, entities/joseph-bernstein, entities/clayton-christensen, entities/adam-smith, entities/dina-powell-mccormick, entities/ilhan-scheer, entities/samuel-weinbach, concepts/talent-flow-china-us, concepts/data-broker-regulation, concepts/economic-possibilities-for-artificial-intelligence, concepts/ai-and-language-models, concepts/a-vision-of-democratic-ai, concepts/algorithmic-decisionmaking, entities/carnegie-mellon, entities/santa-fe-institute, entities/bank-of-england, entities/ecb, entities/federal-reserve, entities/sais, entities/uw. concepts/inverse-cooking-problem and concepts/inverse-trust-problem remain parked under needs-review/2026-07-10-inverse-cooking-trust-coined-terms.md and were not re-litigated.

Self-audit of this run

  • Broken links moved in the right direction. 243 → 237 distinct targets, 268 → 262 refs. Six resolved by page creation. The referenced-2+ count is unchanged at 25, as expected: none of the created pages was on that list, because all seven were dangling prose references rather than broken wikilinks — which is why a broken-link-only candidate list would have missed every one of them.
  • No broken outbound links introduced. All wikilinks on the seven new pages were checked against the page set after writing; zero unresolved. One was caught and fixed during the run — [[entities/bis]] was written before checking, and resolves to entities/bureau-of-industry-and-security.
  • Style scan clean. All seven pages checked against the banned-header, hyperbole, self-reference, standing-Predictions and date-header lists; zero hits on each.
  • Cap respected. Seven gaps actioned against a 6–10 cap; twenty-one candidates found. Eight web pulls, no crawls.
  • Confidence set honestly. All seven pages are confidence: medium. None reached high; the three-independent-source bar was not met for any, and two depend on sources still queued rather than ingested.

One-line summary

Seven pages created live — legislation/coppa, legislation/cfaa, legislation/minnesota-hf1606, litigation/moody-v-netchoice, litigation/authors-guild-v-openai, concepts/foreign-direct-product-rule and entities/jakub-pachocki — clearing five of the six US-law anchor gaps first identified on 2026-08-17; two primary texts queued for review (INGEST-minnesota-hf1606.md, INGEST-moody-v-netchoice-opinion.md); and one routing decision needs the curator, namely whether the seventeen sources_count: 1 legislation pages should be repaired by lint so they stop surfacing as thin anchors every rotation.

Second pass — 2026-08-20 12:40Z

The daily routine fired a second time on the same calendar day, roughly five hours after the run above closed at 00:56 local. Re-running the full workflow would have re-swept slice 7 and overwritten this report, so the scope was narrowed instead: only the items the morning run itself named as unreached were actioned — the two structural third-consecutive deferrals and the highest-in-degree thin anchor in the slice. No candidate actioned above was re-litigated, and no new rotation slice was opened.

Scanned

No new New Developments Log/ files had landed since the morning run (newest remains 2026-08-19 22:08), so the recent-driven lane was empty and the pass ran entirely off the morning report's own deferred backlog. Broken-link state at the start of this pass: 237 distinct targets across 262 refs, 25 referenced 2+ times.

Gaps actioned (3 of 3 attempted)

New pages created (live)

  • Johns Hopkins University — score 3, on its third consecutive deferral and named in the morning report as structural. "Johns Hopkins" appears on 23 content pages with no page behind it. Built as a distributed-footprint entity page rather than a stub: the Data Science and AI Institute's stated build-out (80 tenure-track faculty, 30 Bloomberg Distinguished Professorship endowed chairs, ~30 scientific staff, 1,000+ students) from ai.jhu.edu; the AI-governance cluster (Gillian K. Hadfield's Bloomberg Distinguished Professorship of AI Alignment and Governance, Seth Lazar and the relocated MINT Lab, Henry Farrell at SAIS); the Center for Health Security's biosecurity position via AI-designed viral genomes and the ScreenDNA letter; Matthew Green's encrypted-reasoning disclosure; APL as a named Genesis Mission project site; and Jared Kaplan's Johns Hopkins affiliation on Scaling Laws for Neural Language Models. 949 words, confidence: medium, 6 sources.

The morning report's discrepancy is resolved. Both the 2026-08-19 and 2026-08-20 reports flagged that entities/johns-hopkins "appears in an earlier scan's actioned list but does not exist on disk." It appears in the 2026-08-17 report under link aliases fixed — three broken targets removed from Gillian K. Hadfield during a rewrite — not under pages created. Nothing was lost and no earlier run mis-reported itself; the page had simply never been claimed as built.

  • P(doom) — score 3, also on its third consecutive deferral and named as structural. Inside My AI Law & Policy Class 23: When Silicon Valley's Effective Altruists Meet Washington's Export Controls (Farahany / Hamilton, November 2025) carried [[concepts/p-doom]] (planned) in its Relationships line, so the page was an explicitly deferred dependency rather than a passing mention. Covers the term's rationalist-community origin and 2023 diffusion after GPT-4; the 2023 AI Impacts Expert Survey figures (mean 14.4%, median 5% over 100 years) and the right-skew they imply; the February 2025 Survey 160 general-population result (mean 35.6%, median 20%, ±9.84pp) which runs well above the researcher surveys; a 28-row table of published individual estimates from LeCun's <0.01% to Yampolskiy's 99.9%+; the underspecification criticism (outcome, horizon and conditioning all unfixed) with the LessWrong and AEI arguments; and the economics literature treating the quantity as a modeling input (Jones 2024, AER: Insights; Growiec & Prettner 2026, Economic Modelling). 1,542 words, confidence: medium, 5 sources.

Pages expanded (live)

  • SCALE Act — Rep. John Moolenaar (H.R. 8306, April 2026) — score 4, named in the morning report as "the highest in-degree of the group and the best single expand candidate" (in-degree 11, 606 words, medium/1, unchanged since 2026-06-06). The page previously carried no bill number, an approximate introduction date, and an open question about the undefined "at scale" qualifier. Now: H.R. 8306, 119th Congress, introduced April 15, 2026, referred to House Foreign Affairs, full title Semiconductor Controls Adjusted to Limit Exports Act; the sponsor's term for the mechanism, rolling technical threshold (RTT); the Commerce–DNI metrics mandate and its three named indicators (total processing power, interconnect bandwidth, memory capacity); and the three constraints operating alongside the threshold — presumption of denial where an export would push adversary-controlled AI hardware above five percent of the total US AI hardware base, certification that US supply chains and domestic demand are not disrupted, and preservation of existing controls including the Entity List for sub-threshold exports. 606 → 1,034 words, sources_count 1 → 4. Every pre-existing fact, citation and typed relationship was preserved; the existing "at scale is open to gaming" objection was kept and answered only in part, per the note below.
  • [[entities/sais]] → [[entities/johns-hopkins]] (2 instances, both on Henry Farrell), with "SAIS" retained in prose as the school's name.

This corrects a standing assumption, and the correction matters more than the one fix. Five consecutive runs recorded zero alias fixes and cited entities/sais, entities/uw, entities/ecb as proof that the remaining referenced-2+ targets are "institution shorthand, not aliases of existing pages." The premise was true and the conclusion was wrong: entities/sais was not an alias of an existing page because the institution page did not exist yet. The fix was to create it. That reframes the rest of the institution-shorthand group — entities/carnegie-mellon, entities/santa-fe-institute, entities/bank-of-england, entities/ecb, entities/federal-reserve, entities/uw — as missing-page candidates that each clear a broken target on creation, not as permanently dead links. They should re-enter scoring as type-3 gaps rather than being skipped as settled.

Graph wiring

Nine inbound links added across eight existing pages: Gillian K. Hadfield, Seth Lazar, Henry Farrell, Encrypted Reasoning Traces, AI Biosecurity and Genesis Mission → entities/johns-hopkins; AI Existential Risk, AGI Timelines, Statement on AI Risk (CAIS) and Eliezer Yudkowsky → concepts/p-doom. The (planned) markers on Inside My AI Law & Policy Class 23: When Silicon Valley's Effective Altruists Meet Washington's Export Controls (Farahany / Hamilton, November 2025) were cleared for both concepts/p-doom and concepts/foreign-direct-product-rule, the latter created by the morning run.

Queued — foundational sources

None. This pass created and expanded pages only; no primary text was pulled or queued.

Authenticity-verification failures

None. One provenance note: congress.gov and the Select Committee press-release page both returned connector errors on repeated Firecrawl attempts. The bill number, introduction date and referral were confirmed through a second search tool against the congress.gov H.R. 8306 text page, and the press release was then retrieved in full from chinaselectcommittee.house.gov — the sponsoring committee's own domain — by direct fetch. Both are the canonical hosts for their document types. No secondary or aggregator copy was substituted, and nothing was written from a search snippet alone.

Source-fidelity notes

  • Send date recorded as event date. The page stated the SCALE Act was introduced "in the week of April 21, 2026." April 21 is the date of the Select Committee's press release; congress.gov gives the introduction as April 15, 2026. This is the first fidelity failure class named in the source-fidelity skill, and it had been sitting in the page since 2026-06-06. Corrected, with both dates now stated and distinguished.
  • A press-release gloss is not statutory text. The committee describes "at scale" as "meaningful production levels which can be defined as at least 25 percent of their annual demand." The wording is the committee's characterization, and the page says so rather than presenting 25 percent as a threshold stated in the bill's operative text.
  • An answered objection is not a deleted objection. The page's existing criticism that the "at scale" qualifier is open to gaming was preserved in full. The 25-percent gloss supplies the volume threshold that objection asked for, and the page notes that it does so without settling how pilot or subsidized production is counted.

Deferred backlog after this pass

Unchanged from the morning report except that concepts/p-doom and entities/johns-hopkins are now closed, and the institution-shorthand broken targets are reclassified upward per the note above. Five slice-7 thin anchors remain unreached: legislation/science-new-golden-age (in-degree 6, 277 words, the thinnest in the slice), legislation/genesis-mission (8, 352), legislation/people-first-chatbot-act (8, 399), legislation/lieu-obernolte-ai-bill (in-degree 3, 516 — lower than the morning report's stated 8), legislation/trump-dpa-determination-2026-10 (in-degree 2, 610 — lower than the stated 8). The two corrected in-degree figures drop those pages below the +1 scoring band and they should be re-scored accordingly next run.

The seventeen sources_count: 1 legislation pages remain a routing decision for the curator, not a gap. government/house-office-of-legislative-counsel was checked and dropped as a false candidate: the single occurrence, on CHAT Act and CHAT Act 2.0, refers to a "legislative-counsel draft" of a bill, not to the House office as an actor.

Self-audit of this pass

  • Broken links moved in the right direction. 237 → 235 distinct targets, 262 → 259 refs; referenced-2+ 25 → 24, with entities/sais cleared by the alias fix.
  • No broken outbound links introduced. All 55 wikilinks across the two new pages were resolved against the page set after writing; zero unresolved.
  • Style scan clean. Banned headers 0, wiki self-reference 0, date-stamped headers 29 (unchanged, none added), standing Predictions 0, header hyperbole 0 across all fourteen touched files. One unprecedented sits in pre-existing text on the Farahany class page inside a paraphrase of the source's own p(win) framing, which house style §10 permits; not introduced here and not altered.
  • Fidelity checklist run on the one revision. No sourced fact dropped, no citation dropped, no unsourced claim added, frontmatter repaired rather than inflated, Snapshot integrity not applicable.
  • Scope respected. Three gaps actioned in a deliberately narrowed second pass, bringing the day's total to ten — the top of the 6–10 daily cap, not beyond it. Seven web calls, no crawls.
  • Confidence set honestly. Both new pages medium; the expanded page held at medium despite reaching four sources, since the fast-decay legislative-status claims rest on a single reading of the bill record.

Second-pass summary

Three gaps closed that the morning run had named as stuck: entities/johns-hopkins and concepts/p-doom created, legislation/scale-act-moolenaar expanded with its bill number, correct introduction date and full provision set. One alias fixed — the first in five runs — and the standing "no aliases remain" finding corrected: six institution-shorthand broken targets are missing pages, not dead links, and should re-enter scoring. One source-fidelity error of the send-date-as-event-date class found and repaired in an existing page.

Categories: Briefings | briefing | gap-scan