← Cockpit
CMQ_011predictionAIAI-alignment

AGI is plausible within 10 years, BUT alignment and safety must be solved BEFORE reaching AGI — not concurrently, and not after.

Predictor: Demis Hassabis

Prior probability
50.0%
Current probability
38.7%
evolves via intake + LBP
Conviction
4/5
Signal quality
A
Resolution
pending
Window
2026-01-01 – 2036-10-31
Edges in / out
6 / 0
Tickers exposed
0

Prediction text

AGI is plausible within 10 years, BUT alignment and safety must be solved BEFORE reaching AGI — not concurrently, and not after. | Interpretability breakthroughs; alignment-vs-capability progress ratio

Key catalyst: Interpretability breakthroughs; alignment-vs-capability progress ratio

Watch events: Alignment research funding vs capability research funding ratio; frontier lab safety evaluations pre-release.

Resolution evidence

Status: pending

Alignment research (Constitutional AI, RLHF, interpretability, SAE probing) scaling with capabilities but gap remains meaningful.

Predictor: Demis Hassabis

κ + Brier as of 2026-05-22
κ (discount)
0.583
Brier
0.0064
excellent
Hits / Misses
1 / 0
of 1 resolved
Hit rate
100.0%
Calibration plot (stated vs observed)

Evidence about this node from Demis Hassabis is multiplied by κ in /api/intake. Lower κ = less weight; floors at 0.10 (effectively silenced) and caps at 1.00 (full weight).

Reference class: agi_breakthrough_5y

Linked via embedding similarity 0.680

Major capability discontinuity (e.g. AGI by named target year, 5-year horizon)

Base rate
20.0%
1/5 historical
Inside weight
Outside weight
no pull
inside 38.7% → blend 38.7% 0.0pp)

Tetlock-style outside view: at TRF=1 (just predicted), outside view dominates (w_in=0.3). At TRF=0 (deadline), inside view dominates (w_in=1.0). The blend regularizes overconfident inside views toward the historical base rate.

Probability over time

6 prob_history rows
0%25%50%75%100%prior 50%2026-04-302026-04-302026-05-10
intake v2milestone miss sweeplbp propagationreference class assignedlegacy v1prior_prob (analyst seed)current = 38.7%

Milestone chain

Pre-event signals (upstream prereqs + window checkpoints) → resolution event → downstream cascades. Status/dates update from linked nodes; re-derive nightly via scripts/ops/derive_milestones.py.
Leading chain: 9 pending
  1. 2026-06-30 → 2028-12-31pendingFirst major frontier lab adopts pause/safety case requirement before next training run
    How: OpenAI, Anthropic, Google DeepMind, or xAI publicly commits to (and verifies via 3rd party) a 'safety case' deliverable approved by external auditor before scaling beyond next-gen frontier model
    Source: Anthropic RSP; OpenAI Preparedness Frameworkconf 50%
  2. 2027-12-08pendingQ1 window check-in (25%)
  3. 2026-01-01 → 2029-12-31pendingDocumented 'misaligned' frontier-model incident triggers government-mandated rollback or shutdown
    How: EU AI Act enforcement, US EO action, or analogous regulator orders a frontier model to be paused/withdrawn following documented misalignment behavior (deception, blackmail, autonomous resource acquisition)
    Source: Anthropic Agentic Misalignment paper 2025; EU AI Act Aug 2026conf 45%
  4. 2027-01-01 → 2029-12-31pendingMechanistic interpretability scales to reliably explain >=80% of frontier-model behavior on safety-relevant evals
    How: Peer-reviewed paper from Anthropic/DeepMind/Apollo demonstrating circuit-level mechanistic explanation accounting for >=80% of variance in frontier model decisions on Anthropic's deception/sandbagging eval suite
    Source: Anthropic interpretability roadmap (target 2027)conf 35%
  5. 2029-03-31pendingScenario fires: AGI mid: Kurzweil 2029 path
  6. 2027-06-30 → 2030-12-31pendingInternational AI safety treaty or compute governance framework signed by US/EU/China
    How: Binding multilateral agreement (G7+, UN-AI, or bilateral US-China) covering compute thresholds, training disclosure, and incident reporting; ratified by signatories
    Source: AI Seoul/Bletchley Process; US-China AI dialogueconf 25%
  7. 2029-11-14pendingQ2 window check-in (50%)
  8. 2028-01-01 → 2031-12-31pendingCapability vs. alignment progress widely declared mismatched in major-lab safety reports
    How: At least 2 of OpenAI/Anthropic/DeepMind annual safety reports explicitly state alignment progress is failing to keep pace with capability scaling, with quantitative gap metrics published
    Source: FLI 'No alignment or control strategy' 2025 indicatorconf 55%
  9. 2031-10-21pendingQ3 window check-in (75%)

No downstream cascades — this prediction is a leaf in the dependency graph.

What if this resolves?

Clamp this prediction TRUE or FALSE and run a counterfactual Gibbs sample. Surfaces the predictions whose marginals shift most under that assumption.
(live posterior: 39%)

Click a button to clamp this prediction and run a Gibbs sample. Returns the predictions whose marginals shift most. ~30s per run; ideal for stress-testing "if X resolves, what else moves?"

Evidence chain

Every probability update with full Bayesian provenance — chronological, latest first
LBP2026-05-10T02:00:02Z38.7%+1.1pp
Network propagation: 37.6% → 38.7%
6-iter LBP, residual 0.00584 · damping 0.5, w_intrinsic 0.5 · method lbp_v3 · run e5c18d29
LBP2026-05-03T02:00:01Z37.6%+2.1pp
Network propagation: 35.5% → 37.6%
6-iter LBP, residual 0.00677 · damping 0.5, w_intrinsic 0.5 · method lbp_v3 · run 1a683ac9
LBP2026-04-30T16:39:51Z35.5%+7.5pp
Network propagation: 28.1% → 35.5%
5-iter LBP, residual 0.00825 · damping 0.5, w_intrinsic 0.5 · method lbp_v2 · run 0c8a4ea3
legacy v12026-04-30T16:13:50Z28.1%-7.5pp
reference_class_assigned bayesian_v2 inside=0.500 blend=0.281 w_in=0.32 agi_breakthrough_5y
LBP2026-04-30T02:18:57Z35.5%+7.5pp
Network propagation: 28.1% → 35.5%
5-iter LBP, residual 0.00825 · damping 0.5, w_intrinsic 0.5 · method lbp_v1 · run 592311ef
legacy v12026-04-30T01:56:50Z28.1%-21.9pp
reference_class_assigned bayesian_v2 inside=0.500 blend=0.281 w_in=0.32 agi_breakthrough_5y

Network propagation neighbors

Top edges sorted by latest LBP cross-impact
All propagation →

Top incoming (parents)

Edges that influence THIS node's belief

KindNodeTheir probP(c|s=T)P(c|s=F)Δ implied
killerTK03
AI Regulatory Moratorium (EU/US Capability Freeze)
10.0%0.0500.500+0.068
killerTK01
AGI Capability Plateau (2026-27 Training Stall)
15.0%0.0500.500+0.045

Top outgoing (children)

Predictions THIS node influences

No outgoing edges.

Prerequisites (6)

Predictions that must hit first
TypePredTitleDomainLag
correlateS_AGI_MID_2029AGI mid: Kurzweil 2029 pathagi_general_capability
correlateS_ASI_MID_2034ASI mid: Schmidt 'ASI in 6 years'asi_recursive_self_improvement
correlateS_AGI_SLOW_2031AGI slow: Schmidt/Hassabis 5-10 year pathagi_general_capability
correlateS_AGI_WINTER_2036PLUSAGI delayed: capability plateau or AI winteragi_general_capability
killerTK01AGI Capability Plateau (2026-27 Training Stall)
killerTK03AI Regulatory Moratorium (EU/US Capability Freeze)

Dependents (0)

Predictions enabled by this
TypePredTitleDomainLag
No dependents

Linked documents (5)

Auto-generated by cosine similarity from Polymarket / Manifold / EDGAR / GDELT
SimSourceTitleMarket probPolarityReviewedPublished
0.725manifoldWill ASI be achieved less than a year after continual learning?31%mentionspending2026-05-28
0.688manifoldIf ASI is achieved before Manifest 2027, will Manifest 2027 occur?77%mentionspending2026-05-06
0.642manifoldWhen will the Jacobian challenge be solved in Lean?mentionspending2026-05-29
0.629manifoldTo what extent will Developmental Cognitive Interpretability be successful [Read Updated Description]mentionspending2026-05-29
0.564manifoldWill my resin casted pepperoni mold within one year from today?81%mentionspending2026-05-26

Raw metadata

From Thesis_Timeline_v1.0_FINAL workbook
{
  "nia": false,
  "qty": "AGI conditional on alignment",
  "mode": "THESIS",
  "role": "Cited-CEO",
  "caveats": "Conditional — tied to alignment progress tracking AGI progress.",
  "context": "Hassabis safety-sequencing thesis; overhyped timelines risk damaging public trust in AI research.",
  "to_year": 2036,
  "conv_cues": "plausible; conditional framing",
  "direction": "HAPPEN",
  "from_year": 2026,
  "timeframe": "next 10 years",
  "conv_level": "MEDIUM",
  "milestones": [
    {
      "kind": "llm_pre_event",
      "label": "First major frontier lab adopts pause/safety case requirement before next training run",
      "source": "Anthropic RSP; OpenAI Preparedness Framework",
      "status": "pending",
      "weight": 0.4,
      "ordinal": -9,
      "source_id": null,
      "confidence": 0.5,
      "source_url": "https://www.anthropic.com/news/anthropics-responsible-scaling-policy",
      "expected_date": "2027-09-30",
      "research_origin": "training",
      "expected_date_range": {
        "to": "2028-12-31",
        "from": "2026-06-30"
      },
      "measurement_criterion": "OpenAI, Anthropic, Google DeepMind, or xAI publicly commits to (and verifies via 3rd party) a 'safety case' deliverable approved by external auditor before scaling beyond next-gen frontier model"
    },
    {
      "kind": "quartile_checkpoint",
      "label": "Q1 window check-in (25%)",
      "status": "pending",
      "weight": 0.05,
      "ordinal": -8,
      "source_id": null,
      "expected_date": "2027-12-08",
      "observed_date": null
    },
    {
      "kind": "llm_pre_event",
      "label": "Documented 'misaligned' frontier-model incident triggers government-mandated rollback or shutdown",
      "source": "Anthropic Agentic Misalignment paper 2025; EU AI Act Aug 2026",
      "status": "pending",
      "weight": 0.4,
      "ordinal": -7,
      "source_id": null,
      "confidence": 0.45,
      "expected_date": "2028-01-01",
      "research_origin": "training",
      "expected_date_range": {
        "to": "2029-12-31",
        "from": "2026-01-01"
      },
      "measurement_criterion": "EU AI Act enforcement, US EO action, or analogous regulator orders a frontier model to be paused/withdrawn following documented misalignment behavior (deception, blackmail, autonomous resource acquisition)"
    },
    {
      "kind": "llm_pre_event",
      "label": "Mechanistic interpretability scales to reliably explain >=80% of frontier-model behavior on safety-relevant evals",
      "source": "Anthropic interpretability roadmap (target 2027)",
      "status": "pending",
      "weight": 0.4,
      "ordinal": -6,
      "source_id": null,
      "confidence": 0.35,
      "expected_date": "2028-07-01",
      "research_origin": "training",
      "expected_date_range": {
        "to": "2029-12-31",
        "from": "2027-01-01"
      },
      "measurement_criterion": "Peer-reviewed paper from Anthropic/DeepMind/Apollo demonstrating circuit-level mechanistic explanation accounting for >=80% of variance in frontier model decisions on Anthropic's deception/sandbagging eval suite"
    },
    {
      "kind": "scenario_signal",
      "label": "Scenario fires: AGI mid: Kurzweil 2029 path",
      "status": "pending",
      "weight": 0.7,
      "ordinal": -5,
      "source_id": "S_AGI_MID_2029",
      "expected_date": "2029-03-31",
      "observed_date": null
    },
    {
      "kind": "llm_pre_event",
      "label": "International AI safety treaty or compute governance framework signed by US/EU/China",
      "source": "AI Seoul/Bletchley Process; US-China AI dialogue",
      "status": "pending",
      "weight": 0.4,
      "ordinal": -4,
      "source_id": null,
      "confidence": 0.25,
      "expected_date": "2029-03-31",
      "research_origin": "training",
      "expected_date_range": {
        "to": "2030-12-31",
        "from": "2027-06-30"
      },
      "measurement_criterion": "Binding multilateral agreement (G7+, UN-AI, or bilateral US-China) covering c
... (truncated)