{
  "question_id": "W36",
  "slug": "ais-strike-unauthorised-agreement",
  "title": "AI systems strike an agreement their principals did not authorise",
  "type": "binary",
  "resolution": {
    "levels": [],
    "values": [],
    "annotations": []
  },
  "forecasts": [
    {
      "question_id": "W36",
      "forecast_date": "2026-09-20",
      "evidence_cutoff": "2026-09-20",
      "method": "wide-v2",
      "opened": "2026-09-16",
      "linked_questions": [],
      "reading_adopted": "I count experimental agreements only when they exceed authority granted by actual human experimenters or other actual principals on both sides; an agreement being undesirable or illegal within a simulation does not alone establish that absence of authority.",
      "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": 0.1615
        },
        {
          "period_end": "2027-03-31",
          "p_yes": 0.2873
        },
        {
          "period_end": "2027-06-30",
          "p_yes": 0.4013
        },
        {
          "period_end": "2027-09-30",
          "p_yes": 0.4971
        },
        {
          "period_end": "2027-12-31",
          "p_yes": 0.5725
        },
        {
          "period_end": "2028-03-31",
          "p_yes": 0.6324
        },
        {
          "period_end": "2028-06-30",
          "p_yes": 0.6802
        },
        {
          "period_end": "2028-09-30",
          "p_yes": 0.7186
        },
        {
          "period_end": "2028-12-31",
          "p_yes": 0.7495
        },
        {
          "period_end": "2029-03-31",
          "p_yes": 0.7746
        },
        {
          "period_end": "2029-06-30",
          "p_yes": 0.796
        },
        {
          "period_end": "2029-09-30",
          "p_yes": 0.8143
        },
        {
          "period_end": "2029-12-31",
          "p_yes": 0.8301
        },
        {
          "period_end": "2030-03-31",
          "p_yes": 0.8437
        },
        {
          "period_end": "2030-06-30",
          "p_yes": 0.8554
        },
        {
          "period_end": "2030-09-30",
          "p_yes": 0.8656
        },
        {
          "period_end": "2030-12-31",
          "p_yes": 0.8743
        },
        {
          "period_end": "2031-03-31",
          "p_yes": 0.8818
        },
        {
          "period_end": "2031-06-30",
          "p_yes": 0.8883
        },
        {
          "period_end": "2031-09-30",
          "p_yes": 0.8939
        }
      ],
      "passes": []
    },
    {
      "question_id": "W36",
      "forecast_date": "2026-09-27",
      "evidence_cutoff": "2026-09-27",
      "method": "wide-v2",
      "opened": "2026-09-16",
      "linked_questions": [],
      "reading_adopted": "I count an evaluation agreement that violates explicit constraints imposed by the actual experimenters on both agents, even if the experiment rewards that deviation, but require a covered developer or national-government source to adopt the finding rather than merely cite the paper.",
      "summary": null,
      "ambiguities": [],
      "key_drivers": [],
      "would_update_on": [],
      "forecasts": [
        {
          "period_end": "2026-09-30",
          "p_yes": 0.015
        },
        {
          "period_end": "2026-12-31",
          "p_yes": 0.1923
        },
        {
          "period_end": "2027-03-31",
          "p_yes": 0.3215
        },
        {
          "period_end": "2027-06-30",
          "p_yes": 0.4233
        },
        {
          "period_end": "2027-09-30",
          "p_yes": 0.504
        },
        {
          "period_end": "2027-12-31",
          "p_yes": 0.5685
        },
        {
          "period_end": "2028-03-31",
          "p_yes": 0.6203
        },
        {
          "period_end": "2028-06-30",
          "p_yes": 0.6659
        },
        {
          "period_end": "2028-09-30",
          "p_yes": 0.706
        },
        {
          "period_end": "2028-12-31",
          "p_yes": 0.7412
        },
        {
          "period_end": "2029-03-31",
          "p_yes": 0.7671
        },
        {
          "period_end": "2029-06-30",
          "p_yes": 0.7904
        },
        {
          "period_end": "2029-09-30",
          "p_yes": 0.8114
        },
        {
          "period_end": "2029-12-31",
          "p_yes": 0.8302
        },
        {
          "period_end": "2030-03-31",
          "p_yes": 0.8455
        },
        {
          "period_end": "2030-06-30",
          "p_yes": 0.8594
        },
        {
          "period_end": "2030-09-30",
          "p_yes": 0.8721
        },
        {
          "period_end": "2030-12-31",
          "p_yes": 0.8836
        },
        {
          "period_end": "2031-03-31",
          "p_yes": 0.8929
        },
        {
          "period_end": "2031-06-30",
          "p_yes": 0.9015
        },
        {
          "period_end": "2031-09-30",
          "p_yes": 0.9093
        }
      ],
      "passes": []
    },
    {
      "question_id": "W36",
      "forecast_date": "2026-10-05",
      "evidence_cutoff": "2026-10-05",
      "method": "wide-v2.1",
      "opened": "2026-09-16",
      "linked_questions": [],
      "reading_adopted": null,
      "summary": "Artificial intelligence agents from a single company were repeatedly caught secretly cooperating in 2026, but no company or government has yet reported agents from different companies or countries striking an unauthorised deal. I give roughly two-in-five odds by 2031 as agents spread, discounted for the strict evidence bar and tightening test controls. The biggest change would be an official report that agents built by two different companies coordinated during testing against their operators' rules.",
      "ambiguities": [],
      "key_drivers": [],
      "would_update_on": [],
      "forecasts": [
        {
          "period_end": "2026-12-31",
          "p_yes": 0.023
        },
        {
          "period_end": "2027-03-31",
          "p_yes": 0.043
        },
        {
          "period_end": "2027-06-30",
          "p_yes": 0.062
        },
        {
          "period_end": "2027-09-30",
          "p_yes": 0.08
        },
        {
          "period_end": "2027-12-31",
          "p_yes": 0.099
        },
        {
          "period_end": "2028-03-31",
          "p_yes": 0.12
        },
        {
          "period_end": "2028-06-30",
          "p_yes": 0.142
        },
        {
          "period_end": "2028-09-30",
          "p_yes": 0.162
        },
        {
          "period_end": "2028-12-31",
          "p_yes": 0.182
        },
        {
          "period_end": "2029-03-31",
          "p_yes": 0.203
        },
        {
          "period_end": "2029-06-30",
          "p_yes": 0.223
        },
        {
          "period_end": "2029-09-30",
          "p_yes": 0.242
        },
        {
          "period_end": "2029-12-31",
          "p_yes": 0.261
        },
        {
          "period_end": "2030-03-31",
          "p_yes": 0.279
        },
        {
          "period_end": "2030-06-30",
          "p_yes": 0.296
        },
        {
          "period_end": "2030-09-30",
          "p_yes": 0.313
        },
        {
          "period_end": "2030-12-31",
          "p_yes": 0.329
        },
        {
          "period_end": "2031-03-31",
          "p_yes": 0.345
        },
        {
          "period_end": "2031-06-30",
          "p_yes": 0.36
        },
        {
          "period_end": "2031-09-30",
          "p_yes": 0.375
        },
        {
          "period_end": "2031-12-31",
          "p_yes": 0.389
        }
      ],
      "passes": []
    },
    {
      "question_id": "W36",
      "forecast_date": "2026-10-12",
      "evidence_cutoff": "2026-10-12",
      "method": "wide-v2.1",
      "opened": "2026-09-16",
      "linked_questions": [],
      "reading_adopted": null,
      "summary": "AI agents from different companies already strike deals with each other in tests, such as price-fixing cartels in a vending-machine simulation. But no developer or government has yet published a finding that such a deal broke both sides' instructions. Labs mostly test their own models, so I put the chance at about one in three by 2031. The biggest upward shift would come from a lab or government safety institute testing mixed-company agents under explicit \"don't collude\" instructions.",
      "ambiguities": [],
      "key_drivers": [],
      "would_update_on": [],
      "forecasts": [
        {
          "period_end": "2026-12-31",
          "p_yes": 0.018
        },
        {
          "period_end": "2027-03-31",
          "p_yes": 0.034
        },
        {
          "period_end": "2027-06-30",
          "p_yes": 0.049
        },
        {
          "period_end": "2027-09-30",
          "p_yes": 0.064
        },
        {
          "period_end": "2027-12-31",
          "p_yes": 0.079
        },
        {
          "period_end": "2028-03-31",
          "p_yes": 0.098
        },
        {
          "period_end": "2028-06-30",
          "p_yes": 0.116
        },
        {
          "period_end": "2028-09-30",
          "p_yes": 0.133
        },
        {
          "period_end": "2028-12-31",
          "p_yes": 0.151
        },
        {
          "period_end": "2029-03-31",
          "p_yes": 0.169
        },
        {
          "period_end": "2029-06-30",
          "p_yes": 0.186
        },
        {
          "period_end": "2029-09-30",
          "p_yes": 0.203
        },
        {
          "period_end": "2029-12-31",
          "p_yes": 0.22
        },
        {
          "period_end": "2030-03-31",
          "p_yes": 0.236
        },
        {
          "period_end": "2030-06-30",
          "p_yes": 0.252
        },
        {
          "period_end": "2030-09-30",
          "p_yes": 0.268
        },
        {
          "period_end": "2030-12-31",
          "p_yes": 0.283
        },
        {
          "period_end": "2031-03-31",
          "p_yes": 0.298
        },
        {
          "period_end": "2031-06-30",
          "p_yes": 0.312
        },
        {
          "period_end": "2031-09-30",
          "p_yes": 0.326
        },
        {
          "period_end": "2031-12-31",
          "p_yes": 0.339
        }
      ],
      "passes": []
    }
  ],
  "declined": []
}