{
  "question_id": "W32",
  "slug": "interpretability-confirms-no-hidden-goals",
  "title": "An independent body confirms a model has no hidden goals",
  "type": "binary",
  "resolution": {
    "levels": [],
    "values": [],
    "annotations": []
  },
  "forecasts": [
    {
      "question_id": "W32",
      "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": 0.001
        },
        {
          "period_end": "2026-12-31",
          "p_yes": 0.005
        },
        {
          "period_end": "2027-03-31",
          "p_yes": 0.01
        },
        {
          "period_end": "2027-06-30",
          "p_yes": 0.0159
        },
        {
          "period_end": "2027-09-30",
          "p_yes": 0.0228
        },
        {
          "period_end": "2027-12-31",
          "p_yes": 0.0306
        },
        {
          "period_end": "2028-03-31",
          "p_yes": 0.0403
        },
        {
          "period_end": "2028-06-30",
          "p_yes": 0.0509
        },
        {
          "period_end": "2028-09-30",
          "p_yes": 0.0623
        },
        {
          "period_end": "2028-12-31",
          "p_yes": 0.0744
        },
        {
          "period_end": "2029-03-31",
          "p_yes": 0.0874
        },
        {
          "period_end": "2029-06-30",
          "p_yes": 0.1011
        },
        {
          "period_end": "2029-09-30",
          "p_yes": 0.1155
        },
        {
          "period_end": "2029-12-31",
          "p_yes": 0.1305
        },
        {
          "period_end": "2030-03-31",
          "p_yes": 0.1462
        },
        {
          "period_end": "2030-06-30",
          "p_yes": 0.1624
        },
        {
          "period_end": "2030-09-30",
          "p_yes": 0.1791
        },
        {
          "period_end": "2030-12-31",
          "p_yes": 0.1964
        },
        {
          "period_end": "2031-03-31",
          "p_yes": 0.2141
        },
        {
          "period_end": "2031-06-30",
          "p_yes": 0.2321
        },
        {
          "period_end": "2031-09-30",
          "p_yes": 0.2506
        }
      ],
      "passes": []
    },
    {
      "question_id": "W32",
      "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": 0.001
        },
        {
          "period_end": "2026-12-31",
          "p_yes": 0.005
        },
        {
          "period_end": "2027-03-31",
          "p_yes": 0.011
        },
        {
          "period_end": "2027-06-30",
          "p_yes": 0.0189
        },
        {
          "period_end": "2027-09-30",
          "p_yes": 0.0287
        },
        {
          "period_end": "2027-12-31",
          "p_yes": 0.0403
        },
        {
          "period_end": "2028-03-31",
          "p_yes": 0.0538
        },
        {
          "period_end": "2028-06-30",
          "p_yes": 0.0689
        },
        {
          "period_end": "2028-09-30",
          "p_yes": 0.0857
        },
        {
          "period_end": "2028-12-31",
          "p_yes": 0.104
        },
        {
          "period_end": "2029-03-31",
          "p_yes": 0.1219
        },
        {
          "period_end": "2029-06-30",
          "p_yes": 0.1412
        },
        {
          "period_end": "2029-09-30",
          "p_yes": 0.1618
        },
        {
          "period_end": "2029-12-31",
          "p_yes": 0.1819
        },
        {
          "period_end": "2030-03-31",
          "p_yes": 0.2016
        },
        {
          "period_end": "2030-06-30",
          "p_yes": 0.2207
        },
        {
          "period_end": "2030-09-30",
          "p_yes": 0.2402
        },
        {
          "period_end": "2030-12-31",
          "p_yes": 0.2592
        },
        {
          "period_end": "2031-03-31",
          "p_yes": 0.2777
        },
        {
          "period_end": "2031-06-30",
          "p_yes": 0.2958
        },
        {
          "period_end": "2031-09-30",
          "p_yes": 0.3134
        }
      ],
      "passes": []
    },
    {
      "question_id": "W32",
      "forecast_date": "2026-10-01",
      "evidence_cutoff": "2026-10-01",
      "method": "wide-v2.1",
      "opened": "2026-09-16",
      "linked_questions": [],
      "reading_adopted": "Only an explicit affirmative conclusion of absence counts (e.g., 'we confirm / conclude with high confidence the model has no hidden goals'), reached by the independent body using interpretability tools itself; 'no evidence found', 'very unlikely/low risk' phrasing, or an external reviewer merely agreeing with a developer's interpretability-based claim does not count.",
      "summary": null,
      "ambiguities": [],
      "key_drivers": [],
      "would_update_on": [],
      "forecasts": [
        {
          "period_end": "2026-09-30",
          "p_yes": 0.001
        },
        {
          "period_end": "2026-12-31",
          "p_yes": 0.0025
        },
        {
          "period_end": "2027-03-31",
          "p_yes": 0.0038
        },
        {
          "period_end": "2027-06-30",
          "p_yes": 0.0055
        },
        {
          "period_end": "2027-09-30",
          "p_yes": 0.0076
        },
        {
          "period_end": "2027-12-31",
          "p_yes": 0.0101
        },
        {
          "period_end": "2028-03-31",
          "p_yes": 0.013
        },
        {
          "period_end": "2028-06-30",
          "p_yes": 0.0165
        },
        {
          "period_end": "2028-09-30",
          "p_yes": 0.0204
        },
        {
          "period_end": "2028-12-31",
          "p_yes": 0.0248
        },
        {
          "period_end": "2029-03-31",
          "p_yes": 0.0297
        },
        {
          "period_end": "2029-06-30",
          "p_yes": 0.035
        },
        {
          "period_end": "2029-09-30",
          "p_yes": 0.0406
        },
        {
          "period_end": "2029-12-31",
          "p_yes": 0.0464
        },
        {
          "period_end": "2030-03-31",
          "p_yes": 0.0523
        },
        {
          "period_end": "2030-06-30",
          "p_yes": 0.0584
        },
        {
          "period_end": "2030-09-30",
          "p_yes": 0.0645
        },
        {
          "period_end": "2030-12-31",
          "p_yes": 0.0707
        },
        {
          "period_end": "2031-03-31",
          "p_yes": 0.0769
        },
        {
          "period_end": "2031-06-30",
          "p_yes": 0.0832
        },
        {
          "period_end": "2031-09-30",
          "p_yes": 0.0894
        }
      ],
      "passes": []
    },
    {
      "question_id": "W32",
      "forecast_date": "2026-10-08",
      "evidence_cutoff": "2026-10-08",
      "method": "wide-v2.1",
      "opened": "2026-09-16",
      "linked_questions": [],
      "reading_adopted": null,
      "summary": "I put this at about 7% by the end of 2031 and near zero this year. No outside evaluator or government institute has ever claimed to have confirmed that a leading AI model has no hidden goals. The top evaluators say openly that today's tools cannot support such a claim. New deals putting outside evaluators inside AI companies make it somewhat more likely in later years. A breakthrough letting interpretability tools reliably read a model's goals would most change this.",
      "ambiguities": [],
      "key_drivers": [],
      "would_update_on": [],
      "forecasts": [
        {
          "period_end": "2026-12-31",
          "p_yes": 0.001
        },
        {
          "period_end": "2027-03-31",
          "p_yes": 0.003
        },
        {
          "period_end": "2027-06-30",
          "p_yes": 0.0055
        },
        {
          "period_end": "2027-09-30",
          "p_yes": 0.0085
        },
        {
          "period_end": "2027-12-31",
          "p_yes": 0.0114
        },
        {
          "period_end": "2028-03-31",
          "p_yes": 0.0149
        },
        {
          "period_end": "2028-06-30",
          "p_yes": 0.0184
        },
        {
          "period_end": "2028-09-30",
          "p_yes": 0.0218
        },
        {
          "period_end": "2028-12-31",
          "p_yes": 0.0252
        },
        {
          "period_end": "2029-03-31",
          "p_yes": 0.0291
        },
        {
          "period_end": "2029-06-30",
          "p_yes": 0.033
        },
        {
          "period_end": "2029-09-30",
          "p_yes": 0.0369
        },
        {
          "period_end": "2029-12-31",
          "p_yes": 0.0407
        },
        {
          "period_end": "2030-03-31",
          "p_yes": 0.0446
        },
        {
          "period_end": "2030-06-30",
          "p_yes": 0.0484
        },
        {
          "period_end": "2030-09-30",
          "p_yes": 0.0522
        },
        {
          "period_end": "2030-12-31",
          "p_yes": 0.056
        },
        {
          "period_end": "2031-03-31",
          "p_yes": 0.0598
        },
        {
          "period_end": "2031-06-30",
          "p_yes": 0.0635
        },
        {
          "period_end": "2031-09-30",
          "p_yes": 0.0673
        },
        {
          "period_end": "2031-12-31",
          "p_yes": 0.071
        }
      ],
      "passes": []
    },
    {
      "question_id": "W32",
      "forecast_date": "2026-10-15",
      "evidence_cutoff": "2026-10-15",
      "method": "wide-v2.1",
      "opened": "2026-09-16",
      "linked_questions": [],
      "reading_adopted": null,
      "summary": "No outside evaluator, and not even the companies that build these systems, has ever claimed that tools for inspecting a model's internals proved it has no hidden goals. Recent incidents have made experts even more careful, so they still say only that they found \"no evidence\". The tools are improving fast and new auditing rules are coming, so the chance rises slowly to about 11% by 2031. A required certification of alignment would change this most.",
      "ambiguities": [],
      "key_drivers": [],
      "would_update_on": [],
      "forecasts": [
        {
          "period_end": "2026-12-31",
          "p_yes": 0.0015
        },
        {
          "period_end": "2027-03-31",
          "p_yes": 0.004
        },
        {
          "period_end": "2027-06-30",
          "p_yes": 0.007
        },
        {
          "period_end": "2027-09-30",
          "p_yes": 0.01
        },
        {
          "period_end": "2027-12-31",
          "p_yes": 0.014
        },
        {
          "period_end": "2028-03-31",
          "p_yes": 0.019
        },
        {
          "period_end": "2028-06-30",
          "p_yes": 0.024
        },
        {
          "period_end": "2028-09-30",
          "p_yes": 0.029
        },
        {
          "period_end": "2028-12-31",
          "p_yes": 0.035
        },
        {
          "period_end": "2029-03-31",
          "p_yes": 0.041
        },
        {
          "period_end": "2029-06-30",
          "p_yes": 0.048
        },
        {
          "period_end": "2029-09-30",
          "p_yes": 0.055
        },
        {
          "period_end": "2029-12-31",
          "p_yes": 0.061
        },
        {
          "period_end": "2030-03-31",
          "p_yes": 0.068
        },
        {
          "period_end": "2030-06-30",
          "p_yes": 0.074
        },
        {
          "period_end": "2030-09-30",
          "p_yes": 0.081
        },
        {
          "period_end": "2030-12-31",
          "p_yes": 0.087
        },
        {
          "period_end": "2031-03-31",
          "p_yes": 0.093
        },
        {
          "period_end": "2031-06-30",
          "p_yes": 0.099
        },
        {
          "period_end": "2031-09-30",
          "p_yes": 0.105
        },
        {
          "period_end": "2031-12-31",
          "p_yes": 0.111
        }
      ],
      "passes": []
    }
  ],
  "declined": []
}