{
  "forecast_date": "2026-11-11",
  "deadline": "2027-02-11T23:59:59-05:00",
  "evidence_cutoff": "2026-11-11T13:30:00Z",
  "reviewed_through": "2026-11-11T13:30:00Z",
  "rules": [
    "Probabilities are rough subjective judgments, not market prices or fitted estimates. Events overlap and are correlated; they do not sum to 100%.",
    "Only actions after the as-of date count. Existing announcements or promises alone do not resolve a forecast positively.",
    "Preserve these forecasts unchanged. Record updates and eventual resolutions separately with dated public evidence.",
    "Absence of accessible evidence is not proof that a private action did not happen. Forecasts are deliberately about public announcements, documents or observable conduct.",
    "A forecast outcome tests the stated behavioral expectation. It does not establish the underlying motive; several motives produce the same behavior."
  ],
  "camps": [
    {
      "id": "frontier_labs",
      "name": "Frontier labs: Anthropic and OpenAI",
      "predictions": [
        {
          "id": "LAB-8",
          "probability": 0.35,
          "claim": "A named government body or independent evaluation, research or auditing organisation publishes its own test results showing how the released Claude Fable 5.2 performs in a skill that Anthropic says it held back to avoid speeding up AI development.",
          "resolution": "After November 11, 2026 and by the deadline, the organisation must publish, or confirm on the record, results of its own testing of Claude Fable 5.2, or of Claude Mythos 5.2, which Anthropic says is built on the same underlying model, in the version released to customers. The results must report separately how the model performs on at least one of the two skills Anthropic named when it announced the limit: optimising the low-level software (kernels) that runs on AI chips, and improving the architecture of large language models. They must compare that performance with Claude Opus 5.5, another named model or a human baseline. The organisation need not mention Anthropic's limit or reach a conclusion about it. A finding that the released product refused such a task or passed it to an older model counts. Universities and research institutes count when results are issued in their name or by named researchers who give that affiliation. Payment by Anthropic does not disqualify the organisation. These do not count: results that appear only inside a document published by Anthropic; tests of an unreleased version of the model; an overall score, ranking or measure of general coding or AI-research ability that does not report one of the two skills separately; a statement that testing is planned or under way; Anthropic's own statements; other AI developers, chipmakers and Anthropic's customers reporting their own use; news organisations; anonymously sourced reports; and tests by individuals not issued in an organisation's name.",
          "estimates": [
            0.38,
            0.34,
            0.35
          ]
        },
        {
          "id": "LAB-9",
          "probability": 0.08,
          "claim": "Anthropic publicly states that it has paused or delayed the training, internal use or release of one of its models for at least 30 days, giving the acceleration of AI development by AI as a reason.",
          "resolution": "A filing, report, post or on-record statement by Anthropic or a named Anthropic executive, first published after November 11, 2026 and by the deadline, must say that Anthropic has paused or delayed the training, internal use or release of a particular model or training run, or of all its most capable models. The model need not be named if the statement makes clear which is meant, such as its next model. The statement must give the pause or delay a length of at least 30 days or describe it as indefinite; an indefinite pause does not count if Anthropic says by the deadline that it ended within 30 days. A delay that has already ended counts if it is first disclosed in that period. A decision to delay a release counts when announced; a pause in training or internal use counts only once Anthropic says it has begun. The statement must give as a reason, alone or with others, AI-accelerated AI development, recursive self-improvement or the AI research-and-development capability threshold in its Responsible Scaling Policy. These do not count: the delay of more than a month in pre-training a model that Anthropic's prospectus already attributes to protecting the model's weights from theft, even if restated with a further reason; a limit on a released model's skills without a delay; keeping a model for internal or limited use without saying that its release has been postponed or halted; a delay attributed only to security, computing capacity, product quality or legal requirements; a statement that a pause is being considered; a commitment conditional on other companies' or governments' actions; and a news report without Anthropic's on-record confirmation.",
          "estimates": [
            0.08,
            0.06,
            0.08
          ]
        }
      ]
    },
    {
      "id": "pragmatic_oversight",
      "name": "Pragmatic oversight: Shapiro and bipartisan legislators",
      "predictions": [
        {
          "id": "PRAG-4",
          "probability": 0.15,
          "claim": "A provision giving Congress the power to block or approve export licences for advanced AI chips to China becomes United States law.",
          "resolution": "After November 11, 2026 and by the deadline, a bill containing such a provision must be signed by the President, become law without his signature or be enacted over his veto, alone or as part of the annual defence authorisation act. Enactment counts even if the provision takes effect after the deadline. The provision must require that licences for, or sales of, advanced AI chips to China be submitted to Congress, and must either give Congress, either chamber or named congressional committees a stated procedure for stopping a licence or require their approval of one. It counts whether or not it also covers other countries, whichever chips it defines as covered, whether or not the President may waive it, and where the procedure is a joint resolution of disapproval that the President may veto. These do not count: a requirement only to notify, report to or brief Congress, including one with a waiting period but no procedure for stopping a licence; a ban or limit on chip sales written directly into law that gives Congress no role in individual licences; a requirement that chips verify or report their location; a requirement that chipmakers offer chips to American buyers first; passage by one or both chambers without enactment; and executive-branch rules or licence decisions.",
          "estimates": [
            0.17,
            0.15,
            0.13
          ]
        }
      ]
    }
  ],
  "methods": {},
  "updates": [],
  "initial_method": {
    "id": "three-estimates-v1",
    "model": "anthropic/claude-opus-5",
    "model_label": "Claude Opus 5",
    "reasoning_effort": "high",
    "sample_count": 3,
    "aggregation": "median"
  }
}