{
  "question_id": "W29",
  "slug": "lab-cannot-verify-newest-models-alignment",
  "title": "A lab says it cannot verify its newest model's alignment itself",
  "type": "binary",
  "resolution": {
    "levels": [],
    "values": [],
    "annotations": []
  },
  "forecasts": [
    {
      "question_id": "W29",
      "forecast_date": "2026-09-20",
      "evidence_cutoff": "2026-09-20",
      "method": "wide-v2",
      "opened": "2026-09-16",
      "linked_questions": [],
      "reading_adopted": null,
      "summary": null,
      "ambiguities": [],
      "key_drivers": [],
      "would_update_on": [],
      "forecasts": [
        {
          "period_end": "2026-09-30",
          "p_yes": 0.003
        },
        {
          "period_end": "2026-12-31",
          "p_yes": 0.018
        },
        {
          "period_end": "2027-03-31",
          "p_yes": 0.0376
        },
        {
          "period_end": "2027-06-30",
          "p_yes": 0.0617
        },
        {
          "period_end": "2027-09-30",
          "p_yes": 0.0898
        },
        {
          "period_end": "2027-12-31",
          "p_yes": 0.1171
        },
        {
          "period_end": "2028-03-31",
          "p_yes": 0.148
        },
        {
          "period_end": "2028-06-30",
          "p_yes": 0.1778
        },
        {
          "period_end": "2028-09-30",
          "p_yes": 0.2066
        },
        {
          "period_end": "2028-12-31",
          "p_yes": 0.2344
        },
        {
          "period_end": "2029-03-31",
          "p_yes": 0.2612
        },
        {
          "period_end": "2029-06-30",
          "p_yes": 0.287
        },
        {
          "period_end": "2029-09-30",
          "p_yes": 0.3084
        },
        {
          "period_end": "2029-12-31",
          "p_yes": 0.3292
        },
        {
          "period_end": "2030-03-31",
          "p_yes": 0.3493
        },
        {
          "period_end": "2030-06-30",
          "p_yes": 0.3688
        },
        {
          "period_end": "2030-09-30",
          "p_yes": 0.3846
        },
        {
          "period_end": "2030-12-31",
          "p_yes": 0.4
        },
        {
          "period_end": "2031-03-31",
          "p_yes": 0.415
        },
        {
          "period_end": "2031-06-30",
          "p_yes": 0.4296
        },
        {
          "period_end": "2031-09-30",
          "p_yes": 0.4439
        }
      ],
      "passes": []
    },
    {
      "question_id": "W29",
      "forecast_date": "2026-09-27",
      "evidence_cutoff": "2026-09-27",
      "method": "wide-v2",
      "opened": "2026-09-16",
      "linked_questions": [],
      "reading_adopted": null,
      "summary": null,
      "ambiguities": [],
      "key_drivers": [],
      "would_update_on": [],
      "forecasts": [
        {
          "period_end": "2026-09-30",
          "p_yes": 0.002
        },
        {
          "period_end": "2026-12-31",
          "p_yes": 0.012
        },
        {
          "period_end": "2027-03-31",
          "p_yes": 0.0238
        },
        {
          "period_end": "2027-06-30",
          "p_yes": 0.0385
        },
        {
          "period_end": "2027-09-30",
          "p_yes": 0.0558
        },
        {
          "period_end": "2027-12-31",
          "p_yes": 0.0766
        },
        {
          "period_end": "2028-03-31",
          "p_yes": 0.1006
        },
        {
          "period_end": "2028-06-30",
          "p_yes": 0.1276
        },
        {
          "period_end": "2028-09-30",
          "p_yes": 0.1555
        },
        {
          "period_end": "2028-12-31",
          "p_yes": 0.185
        },
        {
          "period_end": "2029-03-31",
          "p_yes": 0.2136
        },
        {
          "period_end": "2029-06-30",
          "p_yes": 0.2419
        },
        {
          "period_end": "2029-09-30",
          "p_yes": 0.2692
        },
        {
          "period_end": "2029-12-31",
          "p_yes": 0.2947
        },
        {
          "period_end": "2030-03-31",
          "p_yes": 0.3194
        },
        {
          "period_end": "2030-06-30",
          "p_yes": 0.3426
        },
        {
          "period_end": "2030-09-30",
          "p_yes": 0.3649
        },
        {
          "period_end": "2030-12-31",
          "p_yes": 0.3859
        },
        {
          "period_end": "2031-03-31",
          "p_yes": 0.4055
        },
        {
          "period_end": "2031-06-30",
          "p_yes": 0.424
        },
        {
          "period_end": "2031-09-30",
          "p_yes": 0.4412
        }
      ],
      "passes": []
    },
    {
      "question_id": "W29",
      "forecast_date": "2026-10-05",
      "evidence_cutoff": "2026-10-05",
      "method": "wide-v2.1",
      "opened": "2026-09-16",
      "linked_questions": [],
      "reading_adopted": null,
      "summary": "AI companies already use older AI models to test and grade their newest ones, but none has admitted that its own people cannot check whether the newest model is safe. That admission would be costly, so it most likely comes only once models clearly outperform human researchers, which some labs think could happen within a year or two. The biggest signal would be a lab's safety report on a model that can do its own research.",
      "ambiguities": [],
      "key_drivers": [],
      "would_update_on": [],
      "forecasts": [
        {
          "period_end": "2026-12-31",
          "p_yes": 0.007
        },
        {
          "period_end": "2027-03-31",
          "p_yes": 0.017
        },
        {
          "period_end": "2027-06-30",
          "p_yes": 0.03
        },
        {
          "period_end": "2027-09-30",
          "p_yes": 0.045
        },
        {
          "period_end": "2027-12-31",
          "p_yes": 0.062
        },
        {
          "period_end": "2028-03-31",
          "p_yes": 0.081
        },
        {
          "period_end": "2028-06-30",
          "p_yes": 0.1
        },
        {
          "period_end": "2028-09-30",
          "p_yes": 0.119
        },
        {
          "period_end": "2028-12-31",
          "p_yes": 0.137
        },
        {
          "period_end": "2029-03-31",
          "p_yes": 0.153
        },
        {
          "period_end": "2029-06-30",
          "p_yes": 0.169
        },
        {
          "period_end": "2029-09-30",
          "p_yes": 0.183
        },
        {
          "period_end": "2029-12-31",
          "p_yes": 0.196
        },
        {
          "period_end": "2030-03-31",
          "p_yes": 0.208
        },
        {
          "period_end": "2030-06-30",
          "p_yes": 0.219
        },
        {
          "period_end": "2030-09-30",
          "p_yes": 0.229
        },
        {
          "period_end": "2030-12-31",
          "p_yes": 0.239
        },
        {
          "period_end": "2031-03-31",
          "p_yes": 0.248
        },
        {
          "period_end": "2031-06-30",
          "p_yes": 0.257
        },
        {
          "period_end": "2031-09-30",
          "p_yes": 0.265
        },
        {
          "period_end": "2031-12-31",
          "p_yes": 0.274
        }
      ],
      "passes": []
    },
    {
      "question_id": "W29",
      "forecast_date": "2026-10-12",
      "evidence_cutoff": "2026-10-12",
      "method": "wide-v2.1",
      "opened": "2026-09-16",
      "linked_questions": [],
      "reading_adopted": null,
      "summary": "No AI developer has said that its own staff cannot check whether its newest model is aligned and that it relies on older models to make that judgement. Labs already have older models review their safety assessments, but only as a supplement, and they have pledged to keep humans in control. Fast progress toward AI that improves itself makes such an admission plausible around 2027–2029. A lab declaring that its models can fully automate AI research would raise the odds most.",
      "ambiguities": [],
      "key_drivers": [],
      "would_update_on": [],
      "forecasts": [
        {
          "period_end": "2026-12-31",
          "p_yes": 0.008
        },
        {
          "period_end": "2027-03-31",
          "p_yes": 0.02
        },
        {
          "period_end": "2027-06-30",
          "p_yes": 0.035
        },
        {
          "period_end": "2027-09-30",
          "p_yes": 0.053
        },
        {
          "period_end": "2027-12-31",
          "p_yes": 0.074
        },
        {
          "period_end": "2028-03-31",
          "p_yes": 0.097
        },
        {
          "period_end": "2028-06-30",
          "p_yes": 0.12
        },
        {
          "period_end": "2028-09-30",
          "p_yes": 0.143
        },
        {
          "period_end": "2028-12-31",
          "p_yes": 0.166
        },
        {
          "period_end": "2029-03-31",
          "p_yes": 0.186
        },
        {
          "period_end": "2029-06-30",
          "p_yes": 0.204
        },
        {
          "period_end": "2029-09-30",
          "p_yes": 0.219
        },
        {
          "period_end": "2029-12-31",
          "p_yes": 0.234
        },
        {
          "period_end": "2030-03-31",
          "p_yes": 0.247
        },
        {
          "period_end": "2030-06-30",
          "p_yes": 0.259
        },
        {
          "period_end": "2030-09-30",
          "p_yes": 0.27
        },
        {
          "period_end": "2030-12-31",
          "p_yes": 0.281
        },
        {
          "period_end": "2031-03-31",
          "p_yes": 0.29
        },
        {
          "period_end": "2031-06-30",
          "p_yes": 0.299
        },
        {
          "period_end": "2031-09-30",
          "p_yes": 0.307
        },
        {
          "period_end": "2031-12-31",
          "p_yes": 0.315
        }
      ],
      "passes": []
    }
  ],
  "declined": []
}