{
  "question_id": "W19",
  "slug": "training-run-sabotaged-by-rival",
  "title": "A rival or foreign state sabotages a lab's training run",
  "type": "binary",
  "resolution": {
    "levels": [],
    "values": [],
    "annotations": []
  },
  "forecasts": [
    {
      "question_id": "W19",
      "forecast_date": "2026-09-20",
      "evidence_cutoff": "2026-09-20",
      "method": "wide-v2",
      "opened": "2026-09-16",
      "linked_questions": [],
      "reading_adopted": "I count intentionally poisoned anti-distillation outputs only when qualifying sources establish that they were actually consumed in a covered rival's parameter-training run without that rival's authorization; ordinary access blocking or model downgrading alone does not suffice.",
      "summary": null,
      "ambiguities": [],
      "key_drivers": [],
      "would_update_on": [],
      "forecasts": [
        {
          "period_end": "2026-09-30",
          "p_yes": 0.006
        },
        {
          "period_end": "2026-12-31",
          "p_yes": 0.0408
        },
        {
          "period_end": "2027-03-31",
          "p_yes": 0.0744
        },
        {
          "period_end": "2027-06-30",
          "p_yes": 0.1114
        },
        {
          "period_end": "2027-09-30",
          "p_yes": 0.1469
        },
        {
          "period_end": "2027-12-31",
          "p_yes": 0.1853
        },
        {
          "period_end": "2028-03-31",
          "p_yes": 0.222
        },
        {
          "period_end": "2028-06-30",
          "p_yes": 0.257
        },
        {
          "period_end": "2028-09-30",
          "p_yes": 0.2904
        },
        {
          "period_end": "2028-12-31",
          "p_yes": 0.3224
        },
        {
          "period_end": "2029-03-31",
          "p_yes": 0.3562
        },
        {
          "period_end": "2029-06-30",
          "p_yes": 0.3884
        },
        {
          "period_end": "2029-09-30",
          "p_yes": 0.419
        },
        {
          "period_end": "2029-12-31",
          "p_yes": 0.4481
        },
        {
          "period_end": "2030-03-31",
          "p_yes": 0.4784
        },
        {
          "period_end": "2030-06-30",
          "p_yes": 0.5071
        },
        {
          "period_end": "2030-09-30",
          "p_yes": 0.5342
        },
        {
          "period_end": "2030-12-31",
          "p_yes": 0.5598
        },
        {
          "period_end": "2031-03-31",
          "p_yes": 0.5862
        },
        {
          "period_end": "2031-06-30",
          "p_yes": 0.6111
        },
        {
          "period_end": "2031-09-30",
          "p_yes": 0.6344
        }
      ],
      "passes": []
    },
    {
      "question_id": "W19",
      "forecast_date": "2026-09-27",
      "evidence_cutoff": "2026-09-27",
      "method": "wide-v2",
      "opened": "2026-09-16",
      "linked_questions": [],
      "reading_adopted": "Deliberate anti-distillation poisoning can qualify if a covered rival actually consumes the corrupted material in training without authorizing the interference, but access blocking, ordinary refusals, and announced model substitution alone do not qualify.",
      "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": 0.0319
        },
        {
          "period_end": "2027-03-31",
          "p_yes": 0.0658
        },
        {
          "period_end": "2027-06-30",
          "p_yes": 0.0985
        },
        {
          "period_end": "2027-09-30",
          "p_yes": 0.1301
        },
        {
          "period_end": "2027-12-31",
          "p_yes": 0.1605
        },
        {
          "period_end": "2028-03-31",
          "p_yes": 0.1941
        },
        {
          "period_end": "2028-06-30",
          "p_yes": 0.2263
        },
        {
          "period_end": "2028-09-30",
          "p_yes": 0.2573
        },
        {
          "period_end": "2028-12-31",
          "p_yes": 0.287
        },
        {
          "period_end": "2029-03-31",
          "p_yes": 0.3191
        },
        {
          "period_end": "2029-06-30",
          "p_yes": 0.3497
        },
        {
          "period_end": "2029-09-30",
          "p_yes": 0.379
        },
        {
          "period_end": "2029-12-31",
          "p_yes": 0.4069
        },
        {
          "period_end": "2030-03-31",
          "p_yes": 0.4366
        },
        {
          "period_end": "2030-06-30",
          "p_yes": 0.4647
        },
        {
          "period_end": "2030-09-30",
          "p_yes": 0.4915
        },
        {
          "period_end": "2030-12-31",
          "p_yes": 0.5169
        },
        {
          "period_end": "2031-03-31",
          "p_yes": 0.5435
        },
        {
          "period_end": "2031-06-30",
          "p_yes": 0.5686
        },
        {
          "period_end": "2031-09-30",
          "p_yes": 0.5923
        }
      ],
      "passes": []
    },
    {
      "question_id": "W19",
      "forecast_date": "2026-10-02",
      "evidence_cutoff": "2026-10-02",
      "method": "wide-v2.1",
      "opened": "2026-09-16",
      "linked_questions": [],
      "reading_adopted": "Read literally: a rival developer's deliberate degradation/poisoning of outputs harvested by a distilling covered developer counts only if a qualifying source confirms the altered material was actually used in that named developer's training run; rogue AI-agent incidents without human intent and anonymized victims 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": 0.005
        },
        {
          "period_end": "2027-03-31",
          "p_yes": 0.009
        },
        {
          "period_end": "2027-06-30",
          "p_yes": 0.014
        },
        {
          "period_end": "2027-09-30",
          "p_yes": 0.019
        },
        {
          "period_end": "2027-12-31",
          "p_yes": 0.024
        },
        {
          "period_end": "2028-03-31",
          "p_yes": 0.029
        },
        {
          "period_end": "2028-06-30",
          "p_yes": 0.034
        },
        {
          "period_end": "2028-09-30",
          "p_yes": 0.04
        },
        {
          "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": 0.062
        },
        {
          "period_end": "2029-12-31",
          "p_yes": 0.068
        },
        {
          "period_end": "2030-03-31",
          "p_yes": 0.073
        },
        {
          "period_end": "2030-06-30",
          "p_yes": 0.079
        },
        {
          "period_end": "2030-09-30",
          "p_yes": 0.084
        },
        {
          "period_end": "2030-12-31",
          "p_yes": 0.09
        },
        {
          "period_end": "2031-03-31",
          "p_yes": 0.095
        },
        {
          "period_end": "2031-06-30",
          "p_yes": 0.101
        },
        {
          "period_end": "2031-09-30",
          "p_yes": 0.106
        }
      ],
      "passes": []
    },
    {
      "question_id": "W19",
      "forecast_date": "2026-10-09",
      "evidence_cutoff": "2026-10-09",
      "method": "wide-v2.1",
      "opened": "2026-09-16",
      "linked_questions": [],
      "reading_adopted": null,
      "summary": "No government, prosecutor or major AI company has ever publicly confirmed that a rival company or a foreign state deliberately sabotaged an AI training run. All of the confirmed state activity so far is theft or copying. Victims, attackers and spy agencies all have reasons to stay quiet, so I put the chance at about 2% by the end of 2027 and 11% by the end of 2031. The biggest upward shift would come from China officially accusing American intelligence of sabotaging a named Chinese lab's training.",
      "ambiguities": [],
      "key_drivers": [],
      "would_update_on": [],
      "forecasts": [
        {
          "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": 0.012
        },
        {
          "period_end": "2027-09-30",
          "p_yes": 0.016
        },
        {
          "period_end": "2027-12-31",
          "p_yes": 0.021
        },
        {
          "period_end": "2028-03-31",
          "p_yes": 0.026
        },
        {
          "period_end": "2028-06-30",
          "p_yes": 0.031
        },
        {
          "period_end": "2028-09-30",
          "p_yes": 0.036
        },
        {
          "period_end": "2028-12-31",
          "p_yes": 0.042
        },
        {
          "period_end": "2029-03-31",
          "p_yes": 0.047
        },
        {
          "period_end": "2029-06-30",
          "p_yes": 0.053
        },
        {
          "period_end": "2029-09-30",
          "p_yes": 0.058
        },
        {
          "period_end": "2029-12-31",
          "p_yes": 0.064
        },
        {
          "period_end": "2030-03-31",
          "p_yes": 0.07
        },
        {
          "period_end": "2030-06-30",
          "p_yes": 0.076
        },
        {
          "period_end": "2030-09-30",
          "p_yes": 0.082
        },
        {
          "period_end": "2030-12-31",
          "p_yes": 0.088
        },
        {
          "period_end": "2031-03-31",
          "p_yes": 0.094
        },
        {
          "period_end": "2031-06-30",
          "p_yes": 0.099
        },
        {
          "period_end": "2031-09-30",
          "p_yes": 0.106
        },
        {
          "period_end": "2031-12-31",
          "p_yes": 0.112
        }
      ],
      "passes": []
    },
    {
      "question_id": "W19",
      "forecast_date": "2026-10-16",
      "evidence_cutoff": "2026-10-16",
      "method": "wide-v2.1",
      "opened": "2026-09-16",
      "linked_questions": [],
      "reading_adopted": null,
      "summary": "No government, prosecutor or AI developer has ever publicly confirmed that a rival company or foreign state deliberately sabotaged an AI training run. The known sabotage cases were insiders acting for personal reasons, and recent accusations are about copying or spying. Rising geopolitical stakes, Iran's attacks on tech firms, and more AI-powered hacking make a first confirmed case somewhat more likely over time, giving about 9% by 2031. An official attribution of a real attack on a lab's training would most change this.",
      "ambiguities": [],
      "key_drivers": [],
      "would_update_on": [],
      "forecasts": [
        {
          "period_end": "2026-12-31",
          "p_yes": 0.003
        },
        {
          "period_end": "2027-03-31",
          "p_yes": 0.0065
        },
        {
          "period_end": "2027-06-30",
          "p_yes": 0.01
        },
        {
          "period_end": "2027-09-30",
          "p_yes": 0.0134
        },
        {
          "period_end": "2027-12-31",
          "p_yes": 0.0169
        },
        {
          "period_end": "2028-03-31",
          "p_yes": 0.021
        },
        {
          "period_end": "2028-06-30",
          "p_yes": 0.0251
        },
        {
          "period_end": "2028-09-30",
          "p_yes": 0.0292
        },
        {
          "period_end": "2028-12-31",
          "p_yes": 0.0333
        },
        {
          "period_end": "2029-03-31",
          "p_yes": 0.0381
        },
        {
          "period_end": "2029-06-30",
          "p_yes": 0.0429
        },
        {
          "period_end": "2029-09-30",
          "p_yes": 0.0477
        },
        {
          "period_end": "2029-12-31",
          "p_yes": 0.0525
        },
        {
          "period_end": "2030-03-31",
          "p_yes": 0.0577
        },
        {
          "period_end": "2030-06-30",
          "p_yes": 0.0629
        },
        {
          "period_end": "2030-09-30",
          "p_yes": 0.068
        },
        {
          "period_end": "2030-12-31",
          "p_yes": 0.0732
        },
        {
          "period_end": "2031-03-31",
          "p_yes": 0.0783
        },
        {
          "period_end": "2031-06-30",
          "p_yes": 0.0833
        },
        {
          "period_end": "2031-09-30",
          "p_yes": 0.0884
        },
        {
          "period_end": "2031-12-31",
          "p_yes": 0.0934
        }
      ],
      "passes": []
    }
  ],
  "declined": []
}