{
  "question_id": "W15",
  "slug": "frontier-model-reasons-in-unreadable-form",
  "title": "A frontier model reasons in a form people cannot read",
  "type": "binary",
  "resolution": {
    "levels": [],
    "values": [],
    "annotations": []
  },
  "forecasts": [
    {
      "question_id": "W15",
      "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": 0.04
        },
        {
          "period_end": "2026-12-31",
          "p_yes": 0.2512
        },
        {
          "period_end": "2027-03-31",
          "p_yes": 0.371
        },
        {
          "period_end": "2027-06-30",
          "p_yes": 0.4654
        },
        {
          "period_end": "2027-09-30",
          "p_yes": 0.5402
        },
        {
          "period_end": "2027-12-31",
          "p_yes": 0.6
        },
        {
          "period_end": "2028-03-31",
          "p_yes": 0.648
        },
        {
          "period_end": "2028-06-30",
          "p_yes": 0.6885
        },
        {
          "period_end": "2028-09-30",
          "p_yes": 0.7227
        },
        {
          "period_end": "2028-12-31",
          "p_yes": 0.7518
        },
        {
          "period_end": "2029-03-31",
          "p_yes": 0.7754
        },
        {
          "period_end": "2029-06-30",
          "p_yes": 0.7956
        },
        {
          "period_end": "2029-09-30",
          "p_yes": 0.813
        },
        {
          "period_end": "2029-12-31",
          "p_yes": 0.828
        },
        {
          "period_end": "2030-03-31",
          "p_yes": 0.8409
        },
        {
          "period_end": "2030-06-30",
          "p_yes": 0.852
        },
        {
          "period_end": "2030-09-30",
          "p_yes": 0.8616
        },
        {
          "period_end": "2030-12-31",
          "p_yes": 0.8699
        },
        {
          "period_end": "2031-03-31",
          "p_yes": 0.8771
        },
        {
          "period_end": "2031-06-30",
          "p_yes": 0.8832
        },
        {
          "period_end": "2031-09-30",
          "p_yes": 0.8885
        }
      ],
      "passes": []
    },
    {
      "question_id": "W15",
      "forecast_date": "2026-09-27",
      "evidence_cutoff": "2026-09-27",
      "method": "wide-v2",
      "opened": "2026-09-16",
      "linked_questions": [],
      "reading_adopted": "A dated substantive revision can qualify, but a refreshed timestamp on unchanged pre-opening text cannot; all three conditions must be supported by in-window developer-authored material.",
      "summary": null,
      "ambiguities": [],
      "key_drivers": [],
      "would_update_on": [],
      "forecasts": [
        {
          "period_end": "2026-09-30",
          "p_yes": 0.025
        },
        {
          "period_end": "2026-12-31",
          "p_yes": 0.181
        },
        {
          "period_end": "2027-03-31",
          "p_yes": 0.2957
        },
        {
          "period_end": "2027-06-30",
          "p_yes": 0.3943
        },
        {
          "period_end": "2027-09-30",
          "p_yes": 0.473
        },
        {
          "period_end": "2027-12-31",
          "p_yes": 0.5363
        },
        {
          "period_end": "2028-03-31",
          "p_yes": 0.5873
        },
        {
          "period_end": "2028-06-30",
          "p_yes": 0.6285
        },
        {
          "period_end": "2028-09-30",
          "p_yes": 0.6657
        },
        {
          "period_end": "2028-12-31",
          "p_yes": 0.6958
        },
        {
          "period_end": "2029-03-31",
          "p_yes": 0.7216
        },
        {
          "period_end": "2029-06-30",
          "p_yes": 0.7439
        },
        {
          "period_end": "2029-09-30",
          "p_yes": 0.7631
        },
        {
          "period_end": "2029-12-31",
          "p_yes": 0.7797
        },
        {
          "period_end": "2030-03-31",
          "p_yes": 0.794
        },
        {
          "period_end": "2030-06-30",
          "p_yes": 0.8064
        },
        {
          "period_end": "2030-09-30",
          "p_yes": 0.817
        },
        {
          "period_end": "2030-12-31",
          "p_yes": 0.8262
        },
        {
          "period_end": "2031-03-31",
          "p_yes": 0.8349
        },
        {
          "period_end": "2031-06-30",
          "p_yes": 0.8431
        },
        {
          "period_end": "2031-09-30",
          "p_yes": 0.851
        }
      ],
      "passes": []
    },
    {
      "question_id": "W15",
      "forecast_date": "2026-10-05",
      "evidence_cutoff": "2026-10-05",
      "method": "wide-v2.1",
      "opened": "2026-09-16",
      "linked_questions": [],
      "reading_adopted": null,
      "summary": "OpenAI's newest model, Astra, is reported to loop some of its internal layers, but OpenAI has never confirmed this in its own documents. It has also never said that this hidden processing can't be read by people. Other leading labs are publicly committing to reasoning that people can read. The bar is high: a lab has to say this explicitly about its top model. I put the chance at about 15% by the end of 2027 and 45% by the end of 2031.",
      "ambiguities": [],
      "key_drivers": [],
      "would_update_on": [],
      "forecasts": [
        {
          "period_end": "2026-12-31",
          "p_yes": 0.03
        },
        {
          "period_end": "2027-03-31",
          "p_yes": 0.061
        },
        {
          "period_end": "2027-06-30",
          "p_yes": 0.091
        },
        {
          "period_end": "2027-09-30",
          "p_yes": 0.12
        },
        {
          "period_end": "2027-12-31",
          "p_yes": 0.148
        },
        {
          "period_end": "2028-03-31",
          "p_yes": 0.176
        },
        {
          "period_end": "2028-06-30",
          "p_yes": 0.204
        },
        {
          "period_end": "2028-09-30",
          "p_yes": 0.23
        },
        {
          "period_end": "2028-12-31",
          "p_yes": 0.255
        },
        {
          "period_end": "2029-03-31",
          "p_yes": 0.277
        },
        {
          "period_end": "2029-06-30",
          "p_yes": 0.298
        },
        {
          "period_end": "2029-09-30",
          "p_yes": 0.318
        },
        {
          "period_end": "2029-12-31",
          "p_yes": 0.338
        },
        {
          "period_end": "2030-03-31",
          "p_yes": 0.355
        },
        {
          "period_end": "2030-06-30",
          "p_yes": 0.371
        },
        {
          "period_end": "2030-09-30",
          "p_yes": 0.386
        },
        {
          "period_end": "2030-12-31",
          "p_yes": 0.402
        },
        {
          "period_end": "2031-03-31",
          "p_yes": 0.415
        },
        {
          "period_end": "2031-06-30",
          "p_yes": 0.428
        },
        {
          "period_end": "2031-09-30",
          "p_yes": 0.44
        },
        {
          "period_end": "2031-12-31",
          "p_yes": 0.453
        }
      ],
      "passes": []
    },
    {
      "question_id": "W15",
      "forecast_date": "2026-10-12",
      "evidence_cutoff": "2026-10-12",
      "method": "wide-v2.1",
      "opened": "2026-09-16",
      "linked_questions": [],
      "reading_adopted": null,
      "summary": "OpenAI's newest models probably already do part of their reasoning in internal loops, and Chinese lab Zhipu plans the same. But no developer has yet said in its own documents that this reasoning cannot be read by people, and closed labs keep architectures secret. I put the chance near 3% this year, 12% by end-2027 and about 42% by 2031. An official OpenAI description of the looped design would most change this.",
      "ambiguities": [],
      "key_drivers": [],
      "would_update_on": [],
      "forecasts": [
        {
          "period_end": "2026-12-31",
          "p_yes": 0.025
        },
        {
          "period_end": "2027-03-31",
          "p_yes": 0.046
        },
        {
          "period_end": "2027-06-30",
          "p_yes": 0.07
        },
        {
          "period_end": "2027-09-30",
          "p_yes": 0.096
        },
        {
          "period_end": "2027-12-31",
          "p_yes": 0.122
        },
        {
          "period_end": "2028-03-31",
          "p_yes": 0.146
        },
        {
          "period_end": "2028-06-30",
          "p_yes": 0.17
        },
        {
          "period_end": "2028-09-30",
          "p_yes": 0.193
        },
        {
          "period_end": "2028-12-31",
          "p_yes": 0.216
        },
        {
          "period_end": "2029-03-31",
          "p_yes": 0.237
        },
        {
          "period_end": "2029-06-30",
          "p_yes": 0.258
        },
        {
          "period_end": "2029-09-30",
          "p_yes": 0.278
        },
        {
          "period_end": "2029-12-31",
          "p_yes": 0.297
        },
        {
          "period_end": "2030-03-31",
          "p_yes": 0.315
        },
        {
          "period_end": "2030-06-30",
          "p_yes": 0.332
        },
        {
          "period_end": "2030-09-30",
          "p_yes": 0.349
        },
        {
          "period_end": "2030-12-31",
          "p_yes": 0.365
        },
        {
          "period_end": "2031-03-31",
          "p_yes": 0.379
        },
        {
          "period_end": "2031-06-30",
          "p_yes": 0.393
        },
        {
          "period_end": "2031-09-30",
          "p_yes": 0.406
        },
        {
          "period_end": "2031-12-31",
          "p_yes": 0.419
        }
      ],
      "passes": []
    }
  ],
  "declined": []
}