Long Yi
← Back to all writing

p(doom) evidence — Node 9: multilateral governance

July 5, 2026

Predictionp(doom)evidencenode-9

← Evidence index · 中文 · Main post

Each section: Claim · Why · Evidence · Analogue · Would update if · Conf (H/M/L).

Parent: node9 multilateral governance
Purpose: Claim | Why | Evidence | Analogue | Would update if | Conf — for every Node 9 probability.

1. Node definition — non-US governance is load-bearing

Claim: By ~2030, multilateral forums + sovereign-AI blocs (EU, India, China diplomacy, Gulf, UN) shape frontier AI at least as much as US federal law on deployment; on training pause, all blocs fail except low-P EU friction tail.

Why: Paris 2025 split (US unsigned, China signed); India 2026 91-party declaration; EU AI Act extraterritorial market access; US playbook shows zero federal training cap. US-centric Node 1–6 underweights these.

Evidence:

  • (internal note) — four-layer forum network
  • (internal note) §21 — treaty tail thin
  • (internal note) — acceleration + soft safety mainstream

Analogue: GDPR Brussels effect on US tech — compliance for market access without US statute.

Would update if: US signs Paris/New Delhi-style multilateral declaration with verification annex.

Conf: H

2. Branch overview — M1 modal 0.58

Claim: Modal (~58%) path = continuous summit diplomacy (Bletchley→Seoul→Paris→India→Geneva 2027), GPAI/OECD soft law, US–China guardrails restart, UN Global Dialogue — no verified training limits, no durable pause.

Why: Every summit since Bletchley widened signatories while narrowing binding content. Paris shifted from safety → investment; India 2026 → sovereign AI / access. Verification R&D immature (Node 3 §21).

Evidence:

  • Summit series table — research_international_ai_governance_platforms.md §2.8
  • Node 4 composite: P(durable multilateral pause) 0.02–0.05
  • Anthropic RSI (2026-06): pause only if multilateral verification — doesn’t exist

Analogue: Climate COP declarations — rising ambition text, flat binding emissions.

Would update if: Geneva 2027 produces mandatory weights-audit protocol adopted by US+CN.

Conf: M

3. P = 0.58 — Modal forum talk (primary branch)

Claim: P(governance resolves to forum talk + voluntary frameworks by 2030) = 0.58.

Why: Default of institutional drift: 160+ evidenced sections across Nodes 1–6 show oversight without halt; multilateral track record weaker than US (Paris unsigned). 58% = slight majority after reserving 14% EU bind, 10% empty-shell, 18% partial deployment, 3% war.

Evidence:

  • (internal note) — modal governance branch weight 58% (master stitch)
  • (internal note) — Bletchley→Seoul→Paris trajectory
  • CFR: The World Is Trying to Govern AI

Analogue: Nuclear test-ban took decades; AI summits are year 2 of diplomacy.

Would update if: UN Scientific Panel 2026 report triggers emergency GA resolution with enforcement fund.

Conf: M

4. Paris 2025 — US/UK refuse; China signs

Claim: Paris AI Action Summit (2025-02) is the watershed for governance fragmentation: ~60 signatories including China, India, UAE; US and UK unsigned.

Why: JD Vance speech — anti-overregulation, don’t partner with “authoritarian” regimes; UK cited lack of national-security clarity. China signed inclusive/sustainable frame — diplomatic win, zero verification cost.

Evidence:

Analogue: Kyoto Protocol — US non-participation didn’t stop others signing weak targets.

Would update if: Second Trump term reverses and signs successor declaration with trade leverage.

Conf: H

5. Seoul 2024 — China skips ministerial

Claim: China attended Seoul 2024 but did not sign ministerial declaration (27 countries + EU); Zhipu signed Frontier AI Safety Commitments corporately.

Why: Prefigured Paris split — China picks summits where development frame dominates; skips safety-coalition text that implies liability or slowdown.

Evidence:

Analogue: China in WTO — engage on terms that preserve sovereignty narrative.

Would update if: China signs next Seoul-style ministerial with US co-sign.

Conf: H

6. India AI Impact 2026 — New Delhi Declaration (anti-freeze frame)

Claim: India AI Impact Summit (2026-02) produced New Delhi Declaration with 88→91 signatories including US and China; frame = AI for All, sovereign AI, democratizing compute — explicitly not development freeze.

Why: PM Modi theme “Sarvajan Hitaya”; seven “chakras” emphasize access, growth, energy — no training moratorium language. Global South agenda-setter against pause coalition. Pakistan/Taiwan absent (diplomatic sensitivities).

Evidence:

Analogue: NIEO 1970s — technology transfer rhetoric, not capacity restraint.

Would update if: Declaration amended with verifiable compute ceiling article.

Conf: M–H

7. UN Global Dialogue on AI — Geneva Jul 2026

Claim: Inaugural UN Global Dialogue (2026-07-06/07, Geneva) will produce shared understanding, not enforcement — consistent with GDC non-regulatory mandate.

Why: A/RES/79/325 establishes inclusive forum; explicitly not treaty body. Co-chairs El Salvador + Estonia signal Global South + digital-state bridge. Back-to-back with ITU AI for Good — standards/development flavor.

Evidence:

  • UN Global Dialogue site
  • research_international_ai_governance_platforms.md §2.1 — GDC → Dialogue
  • research_un_ai_four_mechanisms.md (referenced in platforms doc)

Analogue: WSIS — decade-long process, soft outcomes first.

Would update if: Outcome document includes mandatory national reporting on >10²⁵ FLOP training runs.

Conf: M

8. EU AI Act — GPAI obligations from Aug 2025

Claim: Chapter V GPAI provider obligations legally in force from 2025-08-02; systemic-risk models (>10²⁵ FLOPs) face adversarial testing, incident reporting, cybersecurity duties.

Why: First binding frontier-model regime globally — applies to any provider placing GPAI on EU market regardless of HQ. Open-source exception except systemic-risk models (Art 53).

Evidence:

Analogue: GDPR applicability to US firms — extraterritorial compliance.

Would update if: Digital Omnibus delays Aug 2026 enforcement (political agreement May 2026 said proceeds).

Conf: H

9. P = 0.14 — EU Brussels effect binds US frontier labs

Claim: P(by 2028, ≥1 US frontier lab materially constrained by EU GPAI enforcement — fines, market withdrawal threat, systemic-risk notification) = 0.14.

Why: 14% = tail not modal: US labs can geo-fence EU products; compliance cost ≠ training stop. But Aug 2026 Commission powers (up to 3% turnover fines) + first-mover enforcement could bite OpenAI/Anthropic/Google API/deployment in EU. Not 0.40 (that’s any enforcement action — subset binds US HQ labs).

Evidence:

  • CSA research note — enforcement Aug 2026 confirmed in Digital Omnibus May 2026
  • Node 4 §25: EU formal enforcement P=0.40; member-state licensing tail 0.45
  • Node 4 tail-gov: EU-heavy mandatory licensing P=0.35

Analogue: Apple DMA gatekeeper compliance — costly but not exit.

Would update if: OpenAI/Anthropic publicly restructure EU offerings before Aug 2026 to avoid systemic-risk classification.

Conf: M

10. EU training cap — stays low (0.06)

Claim: P(EU training FLOP cap with enforcement by 2030) = 0.06.

Why: Paris co-chairs France+India chose investment frame; EU AI Act deployment-centric; von der Leyen competitiveness agenda. Caps appear in academic pause plans, not Commission texts.

Evidence:

  • Node 4 §25 — EU training cap P=0.06 modal
  • Paris 2025 — no pause language
  • research_ai_pause_advocacy_playbook.md — EU not pause leader

Analogue: EU chemical law — hazard use restrictions, not ban on chemistry research.

Would update if: Post-incident EU emergency regulation with compute ceiling.

Conf: M

11. P = 0.10 — Empty-shell treaty (dangerous)

Claim: P(global signed AI governance instrument with ≥80 parties but no verification/enforcement body operational by 2030) = 0.10.

Why: Paris + New Delhi already match profile — broad signatories, voluntary pillars, no audit mandate. Danger: policymakers/labs cite signatures as “we’re governed” while race continues. Hinton explicitly warns advisory bodies ≠ restrictions.

Evidence:

Analogue: Paris Agreement early NDCs — pledged without enforced compliance.

Would update if: Signatories fund IAEA-analog with inspection rights and US joins.

Conf: M

12. P = 0.18 — Partial deployment standards (real enforcement)

Claim: P(binding deployment rules — incident reporting, documentation, market withdrawal — without training cap, enforced on frontier providers) = 0.18.

Why: EU AI Act already partial realization; Council of Europe CETS 225 (2024) adds treaty layer for signatories; UK sectoral regulation + AISI evals inform practice. 18% = path where Brussels effect + CoE + AISI network compound — still not pause.

Evidence:

Analogue: Basel III — heavy reporting, no ban on banking.

Would update if: US federal law preempts extraterritorial EU duties (unlikely).

Conf: M

13. P = 0.05 — Verified training-limit treaty

Claim: P(bilateral or multilateral training FLOP ceiling + verification treaty by 2030) = 0.05 (range 0.03–0.08).

Why: Same verification gap as Nodes 3–4. HEM / compute metering immature; US–China agenda mismatch (chips vs safety); India anti-freeze bloc. 5% = extreme tail requiring crisis + tech breakthrough in verification.

Evidence:

  • Node 3 §13 — training ceiling P=0.04–0.06
  • arxiv 2604.04712 — treaty-grade metering immature (Node 4 §19)
  • research_ai_pause_advocacy_playbook.md §1.2 — 7–9 prerequisites missing

Analogue: INF Treaty — required parity + verification tech + political window.

Would update if: US–China guardrails restart produces pilot HEM with public results.

Conf: M–H

14. P = 0.50 — US–China guardrails talks (not pause)

Claim: P(formal US–China AI dialogue restarts and continues through 2028 on guardrails/misuse) = 0.50; P(mutual training freeze | talks) <0.02.

Why: 2026-05 Beijing summit restart commitment; Bessent non-state-actor frame; Track II >8 meetings vs Track I 1 session ever. Talks decouple from pause — CN conditions on chip controls.

Evidence:

  • (internal note) — full inventory
  • Node 3 §20 — P(guardrails | modal)=0.50
  • Node 4 §27 — guardrails P=0.35 modal, 0.40 tail-gov

Analogue: US–SU hotline — crisis comms, not arms abolition.

Would update if: Second Geneva session publishes joint safety standard.

Conf: M

15. P = 0.03 — Taiwan hot war coupling (<5%)

Claim: P(kinetic great-power clash primarily triggered or escalated via AI weight theft/export-control crisis, by 2030) = 0.03.

Why: Node 3 T1 0.02–0.04; AI 2027 kinetic strike on PRC datacenters is scenario device. Stuxnet-class rarity; nuclear shadow; economic interdependence. 3% = upper-mid of Node 3 range, <5% as scoped.

Evidence:

  • Node 3 §11 — T1 kinetic P=0.02–0.04
  • AI 2027 security forecast — discussed strike options
  • CFR cyber ops tracker — few kinetic cyber responses

Analogue: 1962 near-misses — possible, not modal.

Would update if: Documented US offensive cyber on CN training cluster + PRC military retaliation.

Conf: L

16. India sovereign AI — Yotta / BHASHINI / IndiaAI Mission

Claim: P(India operational sovereign-AI stack at population scale — indigenous cloud + GPU + gov platforms — by 2028) = 0.55.

Why: Yotta migrated BHASHINI to Shakti Cloud; Maha Kumbh 2025 stress test; IndiaAI Mission ₹10,000+ crore compute pipeline; pre-summit Sovereign AI Dialogues. 55% = more likely than not on infra, not on frontier parity.

Evidence:

Analogue: India Stack / UPI — sovereign digital public infra at scale.

Would update if: Major outage exposes sovereign cloud cannot host frontier-class training.

Conf: M

17. China — open-weights diplomacy

Claim: P(China uses open-weight frontier releases (DeepSeek-class) as diplomatic tool to frame US as gatekeeper, 2025–2028) = 0.70.

Why: DeepSeek Jan 2025 global shock; WAIC + Track II narratives on developing-country access; Paris/India signings align with inclusive open models language. Compresses US lead → US more paranoid, less willing to pause.

Evidence:

  • Node 4 §27 — DeepSeek → accelerate not pause
  • Carnegie–CISS 2025 dialogue on DeepSeek diffusion — research_us_china_ai_dialogues.md
  • Paris statement — “open AI models in compliance with countries frameworks”

Analogue: Huawei 5G — technology + narrative export.

Would update if: CCP restricts open release of next-gen models domestically.

Conf: M

18. China — race narrative dominates pause

Claim: P(CCP official narrative post-alignment scare = “cannot pause unilaterally” / race with US) = 0.75 conditional on global salience event.

Why: Every public CN intervention (Paris speech, WAIC) pairs safety principles with development rights. Geneva 2024: CN delegation focused on lifting chip controls. No CN signature on Seoul ministerial or REAIM Blueprint.

Evidence:

  • Node 4 §27 — CCP public comment P=0.30 modal ( race frame in table)
  • AI 2027 — “two months behind” → accelerate
  • REAIM: CN didn’t sign Blueprint — research_international_ai_governance_platforms.md §2.9

Analogue: PRC nuclear doctrine — minimum deterrence while modernizing.

Would update if: State Council white paper proposes coordinated training moratorium with verification.

Conf: M

19. Actor — EU / AI Office

OutcomeP(modal)P(tail-gov)Notes
Formal GPAI enforcement action by 20280.400.55Aug 2026 powers
Brussels effect binds US lab deployment0.100.25Subset of enforcement
EU training cap0.060.10
Coordinated EU–US verification treaty0.040.08Node 4 falsifier

Evidence: §8–10 above; Node 4 §25.

Conf: M

20. Actor — China

OutcomeP(modal)P(tail-gov)
Open-weights diplomacy continues0.700.55
Signs empty-shell summit declaration0.850.90
US–China guardrails track active0.500.60
Mutual training freeze<0.02<0.03
REAIM / military AI binding rules0.050.08

Evidence: §14, §17–18; research_us_china_ai_dialogues.md.

Conf: M

21. Actor — India

OutcomeP(modal)P(tail-gov)
Sovereign AI infra at scale (Yotta/BHASHINI)0.550.50
Population-scale sovereign usage (500M+ interactions/yr)0.500.45
Leads anti-freeze Global South coalition0.800.70
Joins verified training-limit treaty0.030.05
Frontier AI Commitments → binding domestic law0.120.20
New Delhi Declaration → binding enforcement0.080.10
DPDP constrains training on Indian personal data0.350.40
Follow-on summit 100+ signatories, no verification0.750.70

Evidence: §6, §16, §51–54.

Conf: M

22. Actor — US (multilateral posture)

OutcomeP(modal)P(tail-gov)
Refuses broad summit declarations0.750.40
Participates guardrails bilaterally0.550.65
Signs empty-shell for diplomatic gain0.150.25
Binds to EU GPAI as de facto standard0.120.30

Evidence: Paris unsigned; New Delhi signed (exception — development frame acceptable); Trump EO voluntary framework.

Conf: M

23. Actor — UK / AUKUS-adjacent / Five Eyes

OutcomeP(modal)P(tail-gov)
AISI eval sharing (UK–AU MoU, May 2026)0.750.80
UK signs Paris-style declaration0.200.40
UK mandatory training cap0.040.08
AUKUS AI warfare coordination accelerates race0.350.45

Evidence:

  • UK–Australia MoU — 2026-05-25
  • UK AISI rebrand to AI Security Institute (Dec 2025) — cyber/natsec emphasis
  • Seismic UK 2025 — 40% want stop; not organized veto (Node 4 §12)

Conf: M

24. Actor — UAE / Gulf third pole

OutcomeP(modal)P(tail-gov)
Sovereign compute ↑ global frontier FLOPs0.650.70
Paris / New Delhi signatory0.950.95
EU-aligned deployment rules for G42 exports0.250.35
Binds to training-limit treaty0.020.03

Evidence:

  • Paris signatory list includes UAE (#55)
  • G42 Stargate UAE, MGX, Mistral partnership — MEI primer; Scenarica “third pole”
  • Gulf AI policy architecture 2026

Conf: M

25. UN / IAEA-for-AI attempts

Claim: P(operational IAEA-analog for AI with inspection rights by 2030) = 0.08; P(advisory panel + Global Dialogue only) = 0.75.

Why: UN Scientific Panel (40 experts) + Global Dialogue = GDC implementation. Hinton/Bengio/Red Lines call for IAEA-model — no US/CN endorsement of inspections. REAIM separates military; GDC non-military.

Evidence:

  • UN Scientific Panel FAQ
  • Global Call for AI Red Lines — IAEA analogy explicit
  • research_international_ai_governance_platforms.md §2.1 vs §2.9

Analogue: IAEA 1957 — took Near East war scare + US–SU bargain; not yet analog for AI.

Would update if: US + CN jointly propose inspection protocol at UNGA.

Conf: L–M

26. Treaty branch — modal guardrails (embedded in M1)

Claim: P(guardrails archetype dominates — best practices, eval sharing, misuse prevention, no metering) = 0.55 by 2030.

Why: 2026 Beijing restart; GPAI Code of Practice; Seoul Frontier Commitments; AISI network — all explicitly not training limits. Bessent: non-state-actor access focus.

Evidence:

  • research_us_china_ai_dialogues.md §2.3 — Bessent protocol
  • EC GPAI Code of Practice signatory taskforce
  • Node 3 §12 — guardrails P≈0.50

Conf: M

27. Treaty branch — empty-shell vs partial deployment (discriminators)

Claim: Discriminator between empty-shell (0.10) and partial deployment (0.18) = presence of market-access enforcement (fines, withdrawal), not signature count.

Why: Paris/New Delhi = high signatures, zero fines. EU AI Act = fewer “treaty” optics, real Commission powers Aug 2026. CoE convention adds legal layer for Europe.

Evidence: §8–12; Council of Europe CETS 225 open for signature 2024.

Would update if: New Delhi Declaration adds mandatory national regulator with penalty powers.

Conf: M

28. P(treaty) — composite update for my pdoom

Claim:

Treaty metricPoint est.Range
Any signed multilateral AI instrument (≥60 parties)0.85already happened
Empty-shell (no verification)0.100.07–0.15
Partial deployment enforcement0.180.12–0.24
Verified training limits0.050.03–0.08

Why: Separates paper from enforcement. Master model previously folded treaty into Node 3 0.04–0.08 — too narrow (missed Paris/India volume).

Evidence: §11–13; Node 3 §13.

Conf: M

29. P(no pause) — Node 9 delta

Claim: Node 9 reinforces P(no effective pause) 0.82–0.84; EU tail −0.04 local; empty-shell +0.03; India/Gulf +0.02.

Why: Weighted branch math in parent doc. Multilateral anti-freeze (India) + US unsigned > EU deployment friction for training (not deployment).

Evidence:

  • my pdoom crux #5 — P(no pause)=0.82
  • Node 6 — preemption ↑ no pause
  • §6 India anti-freeze

Would update if: New Delhi Declaration binds signatories to training halt with verification fund.

Conf: M

30. p(doom) — Node 9 channels

Claim: Node 9 modal adds +0.3–0.8pp to P(extinction by 2050) vs US-only governance counterfactual; EU tail −0.5–1pp; empty-shell +0.5–1pp; Taiwan +2–4pp if fires.

Why: Speed and coordination channels — not independent extinction causes. Empty-shell worst for alignment timing (false security). Gulf/India ↑ FLOPs. War tail feeds AI-war bucket.

Evidence:

  • Node 3 §23 — geopolitical lifts
  • correlation matrix — P(no pause) ↔ race acceleration
  • my pdoom emergent sim (~17% doom) pt

Would update if: EU enforcement triggers documented ≥3-month frontier training delay industry-wide.

Conf: L–M

31. Cross-node downstream

If M1 modalIf T1 EU bindIf T2 empty-shell
Node 5 open weights ↑ diplomatic pressureNode 4 tail-gov partially realizedNode 4 P(halt) ↓ — false confidence
Node 3 guardrails talk ↑US labs publish EU compliance cardsUN panel cited as “governance”
Node 6 GAAIA US-only fightBrussels parallel to SB 53India hosts next summit, more signatories

Conf: M

32. Falsifiers (watchlist)

ObservationBranch killed / updated
US+CN verified training treatyM1, T2; P(treaty verified) ↑
EU no GPAI fines by 2029T1 ↓
India pause summit coalition§6 anti-freeze wrong
≥2 labs halt >90d citing New DelhiP(no pause) ↓ major
Taiwan kinetic clash AI-triggeredT4 live
Korea AI Basic Act fines US frontier lab§38 ↑; T3 ↑
Japan blocks MATCH alignment§35 ↓
Brazil Marco Legal fails Chamber vote§42 ↓
Indonesia Perpres includes mandatory model registration§46 ↓

33. Japan — AI Promotion Act & light-touch domestic governance

Claim: P(Japan’s AI Promotion Act remains light-touch — no AI-specific fines, voluntary METI/MIC guidelines dominant — through 2030) = 0.75.

Why: Act fully effective Sep 2025; AI Basic Plan (Dec 2025) emphasizes adoption + trust, not EU-style ex ante risk tiers. Only obligation on private sector = cooperate with government initiatives. Cultural/commercial compliance via reputation (“name and shame”) not penalties.

Evidence:

Analogue: Japan’s early internet regulation — industry guidelines before statute.

Would update if: Diet amends AI Promotion Act with EU-style GPAI systemic-risk duties.

Conf: M

34. Japan — Hiroshima AI Process Action Plan 2026

Claim: P(HAIP Friends Group 66 countries operationalize Action Plan 2026 as voluntary guardrails only — no verification body) = 0.85.

Why: Mar 2026 Tokyo meeting explicitly “on a voluntary basis”; Reporting Framework = corporate disclosure not audits. JICA Global South capacity programs ≠ enforcement. Paris 2025 already announced HAIP Reporting Framework — same archetype as Seoul Frontier Commitments.

Evidence:

Analogue: OECD AI Principles adoption — broad uptake, soft compliance.

Would update if: Friends Group funds mandatory third-party model audits with penalty backing.

Conf: M–H

35. Japan — chip alliance & MATCH Act pressure

Claim: P(Japan tightens METI export controls toward US equivalence if MATCH Act passes, materially restricting CN advanced compute by 2028) = 0.55.

Why: 23-item SME controls since Jul 2023; Nov 2025 FPLD addition; Apr 2026 MATCH Act = 150-day ultimatum. METI Vice Minister Matsuo framed Genesis Mission as complementary to Semiconductor Strategy. Tokyo Electron, SCREEN already planning for reduced China revenue. Re-export gap remains but US pressure ↑.

Evidence:

Analogue: Wassenaar Arrangement harmonization after US unilateral moves.

Would update if: Japan negotiates MATCH carve-out preserving autonomous export licensing.

Conf: M

36. P = 0.60 — Japan Genesis / domestic fab ↑ allied FLOPs

Claim: P(US–Japan Genesis Mission + TSMC Kumamoto + Rapidus materially ↑ non-China allied frontier training capacity by 2028) = 0.60.

Why: $1B five-year DOE–MEXT/METI; Rapidus targeting 2nm; Kumamoto Phase 2 expansion reports. Not same as governance pause — accelerates US-allied compute pole. Correlates Node 7.

Evidence:

  • Genesis Mission announcement Jan 2026
  • TIMEWELL semiconductor 2026 column — Kumamoto, Rapidus timeline

Analogue: CHIPS Act allied fab buildout.

Would update if: Rapidus misses 2027 pilot, Genesis appropriations blocked.

Conf: M

37. Actor — Japan

OutcomeP(modal)P(tail-gov)Notes
HAIP voluntary guardrails lead APAC soft law0.850.80§34
Domestic binding deployment fines (EU-style)0.120.20§33
Chip controls align with US (MATCH or equivalent)0.550.65§35
Signs empty-shell summit declaration0.900.92Paris pattern
Verified training-limit treaty0.040.06

Evidence: §33–36; HAIP membership list includes Global South.

Conf: M

38. Korea — AI Basic Act in force Jan 2026

Claim: P(Korea AI Basic Act produces material binding duties on foreign frontier labs serving KR market by 2028) = 0.28 (subset of Node 9 T3 partial deployment).

Why: Comprehensive statute with extraterritorial reach, high-impact AI duties, generative AI labeling — structurally similar to EU. But grace period ≥1 year, MSIT “minimum regulation,” fines deferred except serious harm. 28% = path where grace ends + enforcement crystallizes; lower than EU 0.40 enforcement because industrial-policy dominance.

Evidence:

Analogue: Korea PIPA early years — law on books, phased enforcement.

Would update if: MSIT announces first GPAI-class fine on foreign lab before 2027.

Conf: M

39. P = 0.35 — Korea grace period extends weak enforcement

Claim: P(Korea AI Basic Act grace period extended or first-wave enforcement remains guidance-only through 2027) = 0.35.

Why: Government commentary emphasizes ecosystem development; National AI Committee chaired by President signals industrial priority; historical pattern of tech regulation soft launch (platform laws). Competes with §38 — not mutually exclusive by year.

Evidence:

  • Trade.gov grace period language
  • Littler — phased enforcement emphasis

Analogue: EU AI Act transition periods for SMEs.

Would update if: Corrective order issued to major US lab in 2026.

Conf: M

40. Korea — sovereign AI industrial stack (260k GPU + NACC)

Claim: P(Korea operational sovereign GPU stack at ≥200k installed Blackwell-class GPUs + National AI Computing Center breaking ground by 2028) = 0.65.

Why: APEC Oct 2025 binding announcements with Samsung/Hyundai/SK/Naver; Samsung SDS NACC operator finalized May 2026; 2.5T won investment, 15k chips in NACC alone. 65% discounts supply-chain slip and Nvidia export friction tail.

Evidence:

Analogue: Saudi NEOM compute pledges — high P on announce, lower on full delivery.

Would update if: <50k GPUs installed by end 2027.

Conf: M

41. Actor — South Korea

OutcomeP(modal)P(tail-gov)Notes
Sovereign GPU stack at scale0.650.60§40
AI Basic Act binds foreign labs0.280.40§38
”Third Way” avoids US–CN binary0.500.45Diplomat
Leads APAC deployment-standard bloc0.400.50With EU, not instead
Mutual training freeze w/ US or CN<0.02<0.03

Evidence: §38–40.

Conf: M

42. Brazil — PL 2338 / Marco Legal da IA

Claim: P(Brazil enacts Marco Legal da IA (PL 2338/2023) with risk-based deployment duties by 2028) = 0.55.

Why: Senate approved Dec 2024; Chamber Special Commission active; regional LATAM pressure to legislate. 55% not higher — Chamber may amend heavily, conciliation delay, copyright fights (audiência pública Apr 2026). Not yet law.

Evidence:

Analogue: LGPD enactment — multi-year legislative fight, eventual pass.

Would update if: Chamber rapporteur report rejected or bill withdrawn.

Conf: M

43. P = 0.20 — Brazil deployment rules bind foreign frontier labs

Claim: P(Enacted Marco Legal materially constrains US/EU frontier lab API/deployment in Brazil by 2030) = 0.20.

Why: SIA decentralized enforcement across sector regulators — weaker than EU AI Office centralization. Fines up to R$50M significant but market smaller than EU; labs may geo-limit. 20% = tail within T3.

Evidence:

  • PL 2338 Senate text — excessive/high/non-high risk tiers
  • CMS — no comprehensive statute yet; LGPD applies today

Analogue: Brazil GDPR — compliance for multinationals, not training stop.

Would update if: First fine on OpenAI/Google Brazil entity post-enactment.

Conf: L–M

44. Brazil — LATAM anti-freeze diplomacy

Claim: P(Brazil leads LATAM access / anti-gatekeeper AI diplomacy at multilateral forums, 2025–2028) = 0.70.

Why: EBIA development frame; likely Paris/New Delhi signatory pattern; Lula digital sovereignty rhetoric; PL 2338 allows non-profit training on copyrighted material — pro-access signal. Aligns with India Global South coalition.

Evidence:

  • Regulations.ai — EBIA, CITDigital Decree 12.308/2024
  • Paris/New Delhi signatory lists include Brazil

Analogue: BRICS digital governance statements.

Would update if: Brazil co-hosts pause coalition summit.

Conf: M

45. Indonesia — Perpres AI Ethics + Roadmap 2026

Claim: P(Indonesia promulgates both Perpres on AI Ethics and National AI Roadmap 2026–2029 by end 2026) = 0.70.

Why: Keppres 38/2025 programs 46 Perpres for 2026; Komdigi submitted final drafts to State Secretariat Jun 2026; Minister Meutya Hafid public commitment. 30% failure = Big Tech lobbying + harmonization delay.

Evidence:

Analogue: Indonesia personal data law — long draft, eventual Perpres.

Would update if: Perpres removed from 2026 program in revision.

Conf: M

46. P = 0.55 — US Big Tech weakens Indonesia Perpres enforcement

Claim: P(US-ABC / AmCham lobbying materially weakens mandatory ethics enforcement clauses in final Perpres) = 0.55.

Why: Jun 2026 Jakarta Post reporting — “largely technical” ethics concerns; US stakeholders questioned benchmarking. Indonesia prioritizes investment attraction; Komdigi dual mandate (ethics + growth). 55% = partial weakening not full abandonment.

Evidence:

  • Jakarta Post Jun 2026
  • Regulations.ai — risk-based obligations in draft

Analogue: EU AI Act lobbying outcomes — high-risk categories narrowed.

Would update if: Final Perpres includes mandatory registration + fines for frontier models.

Conf: M

47. African Union — Continental AI Strategy implementation

Claim: P(≥15 AU member states adopt harmonized national AI strategies aligned to Continental Strategy by 2030) = 0.45.

Why: Strategy adopted Jul 2024; 2025–2030 implementation plan; AU technical assistance on Data Policy Framework (22/35 states Apr 2025). Infrastructure/skills gaps limit uptake. 45% = meaningful but not majority of 55 members.

Evidence:

Analogue: AU Agenda 2063 flagship projects — uneven national uptake.

Would update if: 2027 midterm review shows <8 national strategies published.

Conf: M

48. P = 0.08 — AU binding continental AI treaty

Claim: P(AU negotiates binding continental AI treaty with enforcement body by 2030) = 0.08.

Why: Strategy explicitly non-binding; focus on harmonization + investment; AU has no supranational enforcement like EU. 8% = higher than training-limit global treaty because regional scope smaller — still low.

Evidence:

  • Continental Strategy text — policy recommendations not treaty
  • §47 implementation plan

Analogue: AU Malabo Convention — years to ratify, weak enforcement.

Would update if: AU Executive Council mandates REC-level binding AI protocol.

Conf: L–M

49. P = 0.65 — AU Global South access diplomacy at UN

Claim: P(AU coordinates access / anti-gatekeeper interventions at UN Global Dialogue and related forums, 2026–2028) = 0.65.

Why: Strategy pillar “Promote African participation in shaping global AI governance”; development-first framing; aligns with India New Delhi Declaration. Competes with EU deployment-regulation export.

Evidence:

  • AU Strategy focus area xiii–xv — global governance participation
  • India 2026 Global South coalition §6

Analogue: AU climate negotiation bloc — unified rhetoric, diverse implementation.

Would update if: AU breaks with India on open-weights vs safety standards.

Conf: M

50. Actor — Global South composite (Brazil / Indonesia / AU)

OutcomeP(modal)P(tail-gov)Notes
Collective anti-freeze / access diplomacy0.750.65§44, §49
Any GS deployment law w/ foreign-lab enforcement0.220.30BR+ID+IN subset
Empty-shell summit accumulation0.800.75New Delhi pattern
Verified training-limit treaty participation0.040.06
AU+BR+ID block EU Brussels effect0.150.10Too fragmented

Evidence: §42–49.

Conf: M

51. India — New Delhi Declaration enforcement gap

Claim: P(New Delhi Declaration (91 signatories) produces any binding multilateral enforcement mechanism by 2030) = 0.08.

Why: MediaNama, Economic Diplomacy, DFRLab unanimous — voluntary, no reporting, no funding, no fines. Did not reference UN Scientific Panel or Global Dialogue. Dangerous for T2 empty-shell branch — high signatures, zero enforcement.

Evidence:

Analogue: Paris 2025 statement — same archetype, more signatories.

Would update if: Follow-on ministerial creates funded secretariat with audit mandate.

Conf: M–H

52. India — DPDP + IT Rules as real enforcement (vs summit paper)

Claim: P(India DPDP Act + Rules 2025 materially constrain frontier training on Indian personal data by 2028) = 0.35; P(IT Rules 2026 operational takedown/labeling for AI harms) = 0.70.

Why: No standalone AI Act — but DPDP breach penalties up to ₹250 crore; 18-month phased compliance → full bite ~May 2027. IT Amendment 2026 targets deepfakes now. MeitY Feb 2026 Governance Philosophy = principle-based, gap-filling via existing law. 35% for training-data constraint — consent/lawful basis unsettled for web-scale scraping.

Evidence:

Analogue: GDPR effect on US tech — real but narrower than AI Act.

Would update if: Data Protection Board issues landmark AI-training consent ruling.

Conf: M

53. India — Yotta/BHASHINI population-scale sovereign stack

Claim: P(Yotta/BHASHINI sovereign cloud serves ≥500M citizen-facing AI interactions/year by 2028) = 0.50 (revised up from §16 infra-only 0.55 — interaction metric stricter).

Why: Feb 2026 migration complete; Maha Kumbh 2025 stress test; IndiaAI Mission compute pipeline. 50% on usage metric — infra exists but sustained population-scale adoption harder than migration milestone.

Evidence:

Analogue: UPI transaction volume — infra first, adoption follows.

Would update if: Major BHASHINI outage during national event; reversion to hyperscaler.

Conf: M

54. P = 0.75 — India hosts follow-on empty-shell summit (100+ signatories)

Claim: P(India hosts second Global South–leaning AI summit by 2029 with ≥100 signatories, still no verification annex) = 0.75.

Why: India 2026 success (91); diplomatic capital; anti-freeze frame popular. Default institutional drift toward more paper. Competes with Switzerland/security summits — not mutually exclusive.

Evidence:

  • New Delhi Declaration momentum
  • Paris→India summit succession pattern

Analogue: Climate COP host rotation — each host adds pledges, not enforcement.

Would update if: India declines to host; pivots to binding bilateral only.

Conf: M

55. P = 0.22 — Global South deployment rules composite (non-EU enforcement)

Claim: P(≥2 of {Brazil Marco Legal, Indonesia Perpres, Korea AI Basic Act, India DPDP} produce material foreign-lab deployment duties enforced by 2030) = 0.22 — consolidates T3 Global South/East Asia slice.

Why: Korea highest single-actor P(0.28); Brazil/Indonesia lag; India DPDP partial. 22% = correlated uplift to Node 9 T3 0.18 — may justify revising T3 to 0.20–0.24 on next stitch.

Evidence: §38, §42–43, §45–46, §52.

Would update if: Three jurisdictions fine same US lab within 12 months.

Conf: M

Source index