← Evidence index · 中文 · Main post
Each section: Claim · Why · Evidence · Analogue · Would update if · Conf (H/M/L).
Parent: Shared Ci spine
Purpose: Which P’s move together; do not multiply as independent when stitching p(doom).
Date: 2026-07-03
How to use
When updating one probability, check linked rows. Joint tails are smaller than product of marginals only if correlation is positive on bad outcomes — we list direction.
Notation: ↑ = positive correlation (one up → other up); ↓ = negative; ≈ = weak.
Master correlation table
| P / event | Correlates ↑ with | Correlates ↓ with | Cluster |
|---|---|---|---|
| P(no federal pause through 2028) | Race acceleration (N3 T3); GAAIA preemption wins (N6 O1); Open-weight status quo (N5) | Multi-lab halt (N4 tail-gov); EU licensing (N4 tail) | Race |
| P(BMIA / screening by 2027-12) | Insider bio salience; Amodei/screendna coalition; EU Biotech Act | OSTP-only gridlock; GAAIA preemption if it kills state reporting | Bio governance |
| P(Tier 3 before screening | Tier 2) | Open-weight tamper (N5 T5-B); regulatory delay tail (N2 TAIL-B); eval sandbagging | BMIA pass; IGSC v4; function-based SOC | Bio tail |
| P(whistleblower modal | E) | Short media cycle; GAAIA audits no pause; labs continue training | Tail-gov halt; EU fast enforcement | Alignment governance |
| P(race accelerates post-theft | N3) | P(no pause); P(export ctrl expansion); natsec framing | US–China treaty (N3 T2); verification R&D spend | Geopolitics |
| P(GAAIA preemption wins) | Industry lobbying; federal weak standard; P(no pause) +1–3pp | SB 53 enforcement; state transparency; Encode wins | Federalism |
| P(open-weight ban) | EU-only licensing (N5 T5-A); not US federal ban | Meta strategy; China open-weights PR | Open weights |
| P(extinction | misalign tail) | Hard takeoff; deceptive alignment at C10; P(no shutdown) | Hanson slow takeoff; multipolar balance | Misalignment |
| P(concentrated harm ↑) | Node 1 labor modal; open weights; deskilling | — | Societal |
| P(US frontier on triopoly | N7) | Stargate/GW deals; P(no pause); P(race post-theft N3) | Self-build >50% FLOPs; EU full decouple | Compute chokepoint |
| P(cloud KYC / registration by 2028) | BIS IaaS enforcement (N7); export ctrl (N3); CN block | Federal gridlock; industry lawsuit block | Compute chokepoint |
| P(Fable-class API recall repeat) | Natsec framing; P(no pause); deemed-export expansion | Commerce “no more recalls” policy | Kill switch |
| P(EU/India sovereign decouple | N7 T7-C) | Node 3 T4 allied fracture; P(fragmentation) | EU–US joint cloud audit; CADA softened | Sovereignty |
| P(outage >30d frontier delay | N7 T7-A) | Grid/cooling tails; US-EAST-1 concentration | Multi-region failover proven; zero >7d outages 2yr | Infra tail |
| P(HEM pilot | N7 T7-D) | Export-controlled DC; Node 3 T2 treaty prep | HEM mandate on Stargate; FlexHEG product slip | Verification |
| P(DPA DC seizure | N7) | Node 3 DPA; Node 4 executive tail; concentration | Actual seizure attempt; legal block | Gov chokepoint |
| P(US governance paralysis S2) | N6 O3 gridlock; N2 TAIL-B delay; P(no pause); EO-only framework | BMIA on time; GAAIA pass; fed CBRN eval | Governance capacity |
| P(constitutional crisis S3) | O1 preemption attempt; litigation vs SB53; race post-theft | State compliance clarity; multi-lab halt | Federalism crisis |
| P(civil war-scale S4) | N1 concentrated harm | Coordinated pause; BMIA | Societal tail |
| P(CI export gating expands | N8) | N3 export ctrl; Fable recall (N7); P(no pause) | Training halt; cyber arms treaty | Cyber / CI |
| P(offensive cyber surge | salient E) | N3 modal post-theft; natsec framing | Multi-lab halt; guardrails treaty | Cyber / CI |
| P(T8-A AI cyber cascade) | T8-D kinetic discussion; N3 T1 kinetic tail | Resilient OT segmentation; no compound E | Cyber tail |
| P(agentic fraud industrialized) | N1 labor modal; deskilling; N5 open-weight misuse | — | Societal |
| P(open-weight blamed | cyber E) | N5 modal; N5 T5-B | Closed-model Mythos tier dominates narrative | Open weights |
| P(hidden near-miss stays hidden | N10) | Tier 3 before screening; S2 paralysis; anthropic shadow | Mandatory disclosure law; SB53 incident duty enforced | Disclosure |
| P(corp safety hollowing | N11) | P(no pause); P(no shutdown); E4 path | Board crisis → halt; PBC enforcement | Governance hollowing |
| P(RSI locus cloud | N12) | Misalign tail; N7 race speed | Energy cap; Hanson variant | Takeoff geometry |
Clusters (move as blocks)
Cluster A — Acceleration / race (tight ↑)
P(no pause) ↔ P(labs continue after whistleblower) ↔ P(race post-theft) ↔ P(GAAIA preemption)Joint tail example: Whistleblower + no halt + theft + preemption ≈ not P₁×P₂×P₃ — treat as ~3–5× the product of marginal tail probs (correlated).
Cluster B — Physical-layer bio mitigation (tight ↑)
P(BMIA) ↔ P(EU screening) ↔ P(RAISE/SB53 compliance) ↔ P(IGSC upgrade)Partial ↓ with Cluster A: Screening can pass while race continues — not mutually exclusive (modal storyline).
Cluster C — Bio tail risk (tight ↑)
P(Tier 3 before screening) ↔ P(N2 TAIL-B delay) ↔ P(N5 open-weight CBRN cascade) ↔ P(hidden near-miss)Cluster D — Transparency without brakes (modal glue)
P(hearings N1) ↔ P(whistleblower oversight N4 modal) ↔ P(voluntary EO review)↓ with: P(training halt), P(mandatory licensing)
Cluster E — Compute chokepoint (Node 7; tight ↑ with Cluster A)
P(triopoly concentration N7) ↔ P(cloud KYC N7) ↔ P(export ctrl N3) ↔ P(no pause) ↔ P(Stargate/GW build)↓ with: P(EU sovereign decouple T7-C), P(self-build >50% FLOPs), P(HEM pilot T7-D)
Kill switch sub-cluster: P(Fable recall) ↔ P(deemed export on models) ↔ P(CN cloud block) — not ↔ P(training halt) (modal decoupling)
Cluster F — Governance capacity (crosscut US paralysis)
P(paralysis S2) ↔ P(N6 O3) ↔ P(N2 bio delay tail) ↔ P(no pause)
P(paralysis S2) ↓ P(BMIA on schedule) [partial — EU/IGSC can offset]
P(crisis S3) ↔ P(EO preemption) [ambiguous ↔ misalign speed]Stitching: When S2 elevated (>25% scenario mix), apply modifier table in crosscut us governance capacity — don’t multiply P(BMIA) × P(GAAIA) × P(no pause) as independent; overlap with N6 O3 is not additive.
Cluster G — Cyber / CI (Node 8)
P(CI export gating N8) ↔ P(export ctrl N3) ↔ P(Fable recall N7)
P(offensive cyber surge) ↔ P(N3 modal post-theft)
P(open-weight cyber uplift) ↔ P(N5 modal)
P(T8-A cascade) ↔ P(T8-D kinetic discussion) [not execution]
P(T8-B vetting) ↔ P(T8-C surveillance) [emergency bundle]
P(semi espionage) ↔ P(N3 theft attempt)Joint tail: Don’t multiply N8 T8-A × N3 T1 kinetic as independent — joint escalation tail ≈ 3–5× product of marginals when both fire.
Cluster H — Disclosure / governance hollowing (N10–N11)
P(hidden near-miss N10) ↔ P(Tier 3 before screening N2) ↔ P(BMIA delay S2)
P(corp hollowing N11) ↔ P(no pause) ↔ P(no shutdown) ↔ P(E4 no public leak N4)
P(interp prod halt N4) ↓ P(misalign conditional) [partial — tail only]
P(epistemic collapse N2 ext) ↔ P(BMIA delay) ↔ P(S2 paralysis)
P(RSI locus cloud N12) ↔ P(misalign tail weight) ↔ P(N7 race speed)
P(RSI locus bio N12) ↔ P(bio tail weight) ↔ P(N10)Stitching: N10 + S2 jointly move bio branch weight — use my pdoom Phase 2b mixture, not P(N10)×P(S2) independent.
Stitching rule for my pdoom
Don’t: P(modal) × P(BMIA) × P(no pause) as independent.
Do: Assign scenario branch weights (modal / bio-tail / misalignment-tail) that already encode correlation, then conditional P(extinction | branch).
Update log
| Date | Change |
|---|---|
| 2026-07-03 | Initial matrix from Phase 4c critique |
| 2026-07-04 | Node 7 rows + Cluster E (compute chokepoint) |
| 2026-07-04 | Cluster F + governance paralysis rows (crosscut us governance capacity) |
| 2026-07-04 | Node 8 rows + Cluster G (cyber/CI) |
| 2026-07-04 | Cluster H + N10–N12 rows (Phase 2b zero omissions) |