{
  "question_id": "W23",
  "slug": "labs-commit-to-readable-reasoning",
  "title": "Three labs commit to readable reasoning, with outside checks",
  "type": "binary",
  "resolution": {
    "levels": [],
    "values": [],
    "annotations": []
  },
  "forecasts": [
    {
      "question_id": "W23",
      "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": 0.007
        },
        {
          "period_end": "2027-03-31",
          "p_yes": 0.0189
        },
        {
          "period_end": "2027-06-30",
          "p_yes": 0.0366
        },
        {
          "period_end": "2027-09-30",
          "p_yes": 0.0578
        },
        {
          "period_end": "2027-12-31",
          "p_yes": 0.0841
        },
        {
          "period_end": "2028-03-31",
          "p_yes": 0.108
        },
        {
          "period_end": "2028-06-30",
          "p_yes": 0.1312
        },
        {
          "period_end": "2028-09-30",
          "p_yes": 0.1529
        },
        {
          "period_end": "2028-12-31",
          "p_yes": 0.1732
        },
        {
          "period_end": "2029-03-31",
          "p_yes": 0.1914
        },
        {
          "period_end": "2029-06-30",
          "p_yes": 0.2076
        },
        {
          "period_end": "2029-09-30",
          "p_yes": 0.2218
        },
        {
          "period_end": "2029-12-31",
          "p_yes": 0.2351
        },
        {
          "period_end": "2030-03-31",
          "p_yes": 0.2473
        },
        {
          "period_end": "2030-06-30",
          "p_yes": 0.2586
        },
        {
          "period_end": "2030-09-30",
          "p_yes": 0.269
        },
        {
          "period_end": "2030-12-31",
          "p_yes": 0.2785
        },
        {
          "period_end": "2031-03-31",
          "p_yes": 0.2871
        },
        {
          "period_end": "2031-06-30",
          "p_yes": 0.295
        },
        {
          "period_end": "2031-09-30",
          "p_yes": 0.302
        }
      ],
      "passes": []
    },
    {
      "question_id": "W23",
      "forecast_date": "2026-09-27",
      "evidence_cutoff": "2026-09-27",
      "method": "wide-v2",
      "opened": "2026-09-16",
      "linked_questions": [],
      "reading_adopted": "I require prospective class-wide commitments and published independent findings on both obligations, but not a blanket ban on latent computation or a guarantee of perfectly faithful reasoning traces.",
      "summary": null,
      "ambiguities": [],
      "key_drivers": [],
      "would_update_on": [],
      "forecasts": [
        {
          "period_end": "2026-09-30",
          "p_yes": 0.0005
        },
        {
          "period_end": "2026-12-31",
          "p_yes": 0.0065
        },
        {
          "period_end": "2027-03-31",
          "p_yes": 0.0214
        },
        {
          "period_end": "2027-06-30",
          "p_yes": 0.0459
        },
        {
          "period_end": "2027-09-30",
          "p_yes": 0.0793
        },
        {
          "period_end": "2027-12-31",
          "p_yes": 0.1161
        },
        {
          "period_end": "2028-03-31",
          "p_yes": 0.1514
        },
        {
          "period_end": "2028-06-30",
          "p_yes": 0.1854
        },
        {
          "period_end": "2028-09-30",
          "p_yes": 0.2139
        },
        {
          "period_end": "2028-12-31",
          "p_yes": 0.2414
        },
        {
          "period_end": "2029-03-31",
          "p_yes": 0.2642
        },
        {
          "period_end": "2029-06-30",
          "p_yes": 0.2862
        },
        {
          "period_end": "2029-09-30",
          "p_yes": 0.3041
        },
        {
          "period_end": "2029-12-31",
          "p_yes": 0.3215
        },
        {
          "period_end": "2030-03-31",
          "p_yes": 0.3351
        },
        {
          "period_end": "2030-06-30",
          "p_yes": 0.3484
        },
        {
          "period_end": "2030-09-30",
          "p_yes": 0.3614
        },
        {
          "period_end": "2030-12-31",
          "p_yes": 0.3742
        },
        {
          "period_end": "2031-03-31",
          "p_yes": 0.3835
        },
        {
          "period_end": "2031-06-30",
          "p_yes": 0.3928
        },
        {
          "period_end": "2031-09-30",
          "p_yes": 0.4019
        }
      ],
      "passes": []
    },
    {
      "question_id": "W23",
      "forecast_date": "2026-09-29",
      "evidence_cutoff": "2026-09-29",
      "method": "wide-v2.1",
      "opened": "2026-09-16",
      "linked_questions": [],
      "reading_adopted": "A firm, prospective undertaking to preserve human-readable reasoning (e.g., no neuralese / a cap on opaque serial depth for frontier models) plus an explicit no-CoT-optimization pledge counts if outside-verified; method-agnostic 'maintain monitorability' pledges satisfiable by activation monitoring or confessions, and descriptions of current practice, do 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": 0.003
        },
        {
          "period_end": "2027-03-31",
          "p_yes": 0.007
        },
        {
          "period_end": "2027-06-30",
          "p_yes": 0.014
        },
        {
          "period_end": "2027-09-30",
          "p_yes": 0.024
        },
        {
          "period_end": "2027-12-31",
          "p_yes": 0.035
        },
        {
          "period_end": "2028-03-31",
          "p_yes": 0.046
        },
        {
          "period_end": "2028-06-30",
          "p_yes": 0.057
        },
        {
          "period_end": "2028-09-30",
          "p_yes": 0.066
        },
        {
          "period_end": "2028-12-31",
          "p_yes": 0.074
        },
        {
          "period_end": "2029-03-31",
          "p_yes": 0.082
        },
        {
          "period_end": "2029-06-30",
          "p_yes": 0.089
        },
        {
          "period_end": "2029-09-30",
          "p_yes": 0.095
        },
        {
          "period_end": "2029-12-31",
          "p_yes": 0.1
        },
        {
          "period_end": "2030-03-31",
          "p_yes": 0.105
        },
        {
          "period_end": "2030-06-30",
          "p_yes": 0.11
        },
        {
          "period_end": "2030-09-30",
          "p_yes": 0.114
        },
        {
          "period_end": "2030-12-31",
          "p_yes": 0.118
        },
        {
          "period_end": "2031-03-31",
          "p_yes": 0.121
        },
        {
          "period_end": "2031-06-30",
          "p_yes": 0.124
        },
        {
          "period_end": "2031-09-30",
          "p_yes": 0.127
        }
      ],
      "passes": []
    },
    {
      "question_id": "W23",
      "forecast_date": "2026-10-06",
      "evidence_cutoff": "2026-10-06",
      "method": "wide-v2.1",
      "opened": "2026-09-16",
      "linked_questions": [],
      "reading_adopted": null,
      "summary": "I put this at about 9% by the end of 2031, and near zero for 2026. No lab has made a firm promise to keep its models' reasoning readable and not train against it. Even safety advocates propose looser, non-binding schemes, and OpenAI says monitoring need not rely on readable reasoning. Outside auditing is being built quickly, though. The biggest change would be a law or industry standard requiring readable reasoning, with outside verification.",
      "ambiguities": [],
      "key_drivers": [],
      "would_update_on": [],
      "forecasts": [
        {
          "period_end": "2026-12-31",
          "p_yes": 0.002
        },
        {
          "period_end": "2027-03-31",
          "p_yes": 0.005
        },
        {
          "period_end": "2027-06-30",
          "p_yes": 0.009
        },
        {
          "period_end": "2027-09-30",
          "p_yes": 0.014
        },
        {
          "period_end": "2027-12-31",
          "p_yes": 0.02
        },
        {
          "period_end": "2028-03-31",
          "p_yes": 0.026
        },
        {
          "period_end": "2028-06-30",
          "p_yes": 0.033
        },
        {
          "period_end": "2028-09-30",
          "p_yes": 0.039
        },
        {
          "period_end": "2028-12-31",
          "p_yes": 0.045
        },
        {
          "period_end": "2029-03-31",
          "p_yes": 0.051
        },
        {
          "period_end": "2029-06-30",
          "p_yes": 0.056
        },
        {
          "period_end": "2029-09-30",
          "p_yes": 0.061
        },
        {
          "period_end": "2029-12-31",
          "p_yes": 0.065
        },
        {
          "period_end": "2030-03-31",
          "p_yes": 0.07
        },
        {
          "period_end": "2030-06-30",
          "p_yes": 0.074
        },
        {
          "period_end": "2030-09-30",
          "p_yes": 0.078
        },
        {
          "period_end": "2030-12-31",
          "p_yes": 0.081
        },
        {
          "period_end": "2031-03-31",
          "p_yes": 0.084
        },
        {
          "period_end": "2031-06-30",
          "p_yes": 0.088
        },
        {
          "period_end": "2031-09-30",
          "p_yes": 0.09
        },
        {
          "period_end": "2031-12-31",
          "p_yes": 0.093
        }
      ],
      "passes": []
    },
    {
      "question_id": "W23",
      "forecast_date": "2026-10-13",
      "evidence_cutoff": "2026-10-13",
      "method": "wide-v2.1",
      "opened": "2026-09-16",
      "linked_questions": [],
      "reading_adopted": null,
      "summary": "No major AI company has yet firmly promised to keep its models' step-by-step reasoning readable and not train against it, with an outside auditor checking. Momentum is strong: OpenAI says an industry-wide commitment is needed, Google DeepMind is urging one, and outside evaluators are being embedded at labs. But the bar of three companies, firm wording and published audit findings is high. A joint pledge by OpenAI, Anthropic and Google that names an auditor would move this most.",
      "ambiguities": [],
      "key_drivers": [],
      "would_update_on": [],
      "forecasts": [
        {
          "period_end": "2026-12-31",
          "p_yes": 0.002
        },
        {
          "period_end": "2027-03-31",
          "p_yes": 0.009
        },
        {
          "period_end": "2027-06-30",
          "p_yes": 0.02
        },
        {
          "period_end": "2027-09-30",
          "p_yes": 0.034
        },
        {
          "period_end": "2027-12-31",
          "p_yes": 0.05
        },
        {
          "period_end": "2028-03-31",
          "p_yes": 0.065
        },
        {
          "period_end": "2028-06-30",
          "p_yes": 0.079
        },
        {
          "period_end": "2028-09-30",
          "p_yes": 0.093
        },
        {
          "period_end": "2028-12-31",
          "p_yes": 0.105
        },
        {
          "period_end": "2029-03-31",
          "p_yes": 0.116
        },
        {
          "period_end": "2029-06-30",
          "p_yes": 0.125
        },
        {
          "period_end": "2029-09-30",
          "p_yes": 0.133
        },
        {
          "period_end": "2029-12-31",
          "p_yes": 0.14
        },
        {
          "period_end": "2030-03-31",
          "p_yes": 0.147
        },
        {
          "period_end": "2030-06-30",
          "p_yes": 0.153
        },
        {
          "period_end": "2030-09-30",
          "p_yes": 0.158
        },
        {
          "period_end": "2030-12-31",
          "p_yes": 0.163
        },
        {
          "period_end": "2031-03-31",
          "p_yes": 0.168
        },
        {
          "period_end": "2031-06-30",
          "p_yes": 0.172
        },
        {
          "period_end": "2031-09-30",
          "p_yes": 0.176
        },
        {
          "period_end": "2031-12-31",
          "p_yes": 0.179
        }
      ],
      "passes": []
    }
  ],
  "declined": []
}