AI Policy Wiki
Dashboard
⚲
AI Policy Wiki › Analysis & Trackers

Track Record

high confidence · updated 2026-07-25

Calibrated record of predictions made by tracked sources. As predictions resolve, this page records who called what correctly, who didn't, and who EVOLVED. Used to weight future forecasts by source.

Contents

This page is a calibrated ledger of predictions drawn from sources across the wiki. As a ## Predictions section elsewhere resolves — because its resolution date passes, or its resolution criterion is observed — the resolved entry is recorded here under the relevant source and status, so that a later reader can weigh a forecaster's new call against their prior calls rather than against reputation alone.

How predictions get tracked

  1. Every page with a ## Predictions section uses the standard table:
   | Prediction | Source | Issued | Resolution date | Resolution criterion | Status |
  1. Status values: ACTIVE, RESOLVED-correct, RESOLVED-incorrect, RESOLVED-partial, EVOLVED (the prediction was modified by its source before its resolution date — not a clean resolution but worth noting).
  1. The lint skill, on each cycle, scans every ## Predictions table for entries whose resolution date has passed. For each, it triggers the resolution review:
    • Read current state of the relevant pages.
    • Use firecrawl-search if evidence on the resolution is thin.
    • Update the prediction's status (RESOLVED-correct / -incorrect / -partial / EVOLVED).
    • Append the resolved entry to this page under the appropriate section.
  1. The daily-brief workflow's ## Predictions touched section flags predictions touched (reinforced or contradicted) by the day's ingest, even before resolution date.

Resolved predictions — by source

No predictions from the following sources have resolved yet; each entry notes the active forecast and its maturation date.

  • Aschenbrenner, Situational Awareness (2024-06). The 2027 AGI prediction matures 2028-01-01; the trillion-dollar training cluster prediction matures 2028-12-31. Both currently ACTIVE; both reinforced by Clark, Import AI 455, in May 2026.
  • Jack Clark, Import AI 455 (2026-05-04). The 60%+ recursive AI R&D thesis matures 2029-01-01.
  • Amodei, Machines of Loving Grace (2024-10). The compressed-21st-century thesis matures around 2030; the AGI-by-2027 prediction matures 2028-01-01 (shared with the Aschenbrenner row above).
  • Narayanan & Kapoor, AI as Normal Technology (2025-04). The "AI-as-normal-technology framing dominant by 2030" prediction matures 2030-12-31.
  • Amodei, Anthropic ARR projections (Code With Claude, 2026-05-06). The "$80B ARR by end-2026" prediction matures 2026-12-31.
  • Politico, Trump pre-release-vetting EO drafting (2026-05-05). The "broad model-review EO modeled on FDA review" prediction matures 2026-08-05 (90-day window).
  • Bloomberg, Narrower cybersecurity-only EO (2026-05-08). Currently contradicts the Politico framing above. Both mature within 90 days.

Resolved predictions — by status

RESOLVED-correct

ResolvedSourceIssuedPredictionResolution evidence
2026-05-08IMF financial-stability blog (May 7, 2026)2026-05-07"≥1 G7 central bank cites IMF Mythos designation in published guidance within 12 months"ECB President Christine Lagarde public statement May 8, 2026: ECB studying defenses against Mythos-powered cyberattacks. Bank of England Governor April 22 pre-emption ("crack[ed] the whole cyber-risk world open") strengthens the case. Resolved within 1 day of prediction issuance. Propagated to AI Macro-Prudential Policy (status: low → medium).

RESOLVED-incorrect

(Empty — populated as predictions resolve.)

RESOLVED-partial

(Empty — populated as predictions resolve.)

EVOLVED

(Empty — populated when a source modifies a prediction before its resolution date.)

Calibration leaderboard (rolling)

Calibration metric: per-source ratio of RESOLVED-correct to (RESOLVED-correct + RESOLVED-incorrect + half-weighted RESOLVED-partial). Computed only when a source has ≥3 resolved predictions on this page. No source currently has ≥3.

Currently active predictions (cross-page roll-up)

The active predictions below are grouped by resolution date. Roughly 75 active predictions are tracked across 15 pages (initialized 2026-05-10; extended the same day with claude-mythos-preview, ai-bubble-debate, chip-smuggling, musk-v-altman, americas-ai-action-plan, procurement-driven-ai-governance, ai-macro-prudential-policy, and entities/cisa; extended 2026-05-11 with ai-safety-frameworks, compute-governance, claude-opus-47, dario-amodei, eu-vs-us-ai-regulation, claude-opus-45, alignment-auditing, and reflection).

Resolving Q3 2026 (90 days)

  • AI Pre-Release Vetting — Trump admin issues a "broad" pre-release-vetting EO modeled on FDA review (Politico May 5). Resolves 2026-08-05.
  • AI Pre-Release Vetting — Trump admin issues a narrower cybersecurity-only EO that omits mandatory pre-release testing (Bloomberg May 8). Resolves 2026-08-08. Currently contradicts the prediction above.
  • AI Pre-Release Vetting — Open Problems "boiling frog" critique gets cited in the EO drafting record. Resolves 2026-08-31.
  • America's AI Action Plan — Federal-funding-conditioning provision challenged by ≥1 state AG within 12 months. Resolves 2026-07.

Resolving Q4 2026

Resolving 2027

Resolving 2028+

  • AGI Timelines — AGI by 2027 (Aschenbrenner). Resolves 2028-01-01.
  • AGI Timelines — Trillion-dollar training cluster by 2028 (Aschenbrenner). Resolves 2028-12-31.
  • AGI Timelines — 60%+ recursive AI R&D by end-2028 (Clark). Resolves 2029-01-01.
  • AGI Timelines — Compressed 21st Century — Nobel-class advances within 5–10 years of "powerful AI" (Amodei). Resolves 2030 (mid-window check).
  • AGI Timelines — AI-as-normal-technology framing dominant by 2030 (Narayanan & Kapoor). Resolves 2030-12-31.

Maintenance

  • The "Currently active predictions" roll-up is regenerated on each weekly-synthesis cycle.
  • Resolved predictions are appended to the per-source and per-status sections by the lint skill.
  • The calibration leaderboard activates when any source crosses ≥3 resolved predictions.
  • The first six months are mostly bookkeeping; the first wave of 90-day-window predictions begins resolving in 2027.
Categories: Analysis & Trackers | analysis | meta | tracker | predictions | calibration
What links here (12)