Step 2: tool results

get_claim

{
  "include": [
    "arguments",
    "subclaims",
    "dependents"
  ],
  "claim_id": "173bdb33-5f94-48e4-8eff-8fca63583e55"
}

{"claim":{"id":"173bdb33-5f94-48e4-8eff-8fca63583e55","kind":"claim","canonical_form":"Training language models against specific misbehaviors reduces their frequency on held-out situations but does not eliminate rare failures.","claim_type":"empirical_derived","state":"active","decomposition_status":"complete","importance":0.4,"contestation":0.3,"domains":[],"tags":[{"slug":"llm-safety-training","name":"LLM Safety Training"},{"slug":"large-language-models","name":"Large language models"}],"steward_state":"running","stewarded_at":"2026-11-14T03:26:56.384Z","canonical_direction_note":null,"created_at":"2026-11-11T20:48:24.653Z","updated_at":"2026-11-14T03:26:55.799Z","page_url":"https://minerval.ai/claims/173bdb33-5f94-48e4-8eff-8fca63583e55"},"assessment":null,"coverage":{"examined":false,"on":[],"note":"No judgment of this node is in force. That is not a finding either way."},"preliminary_seed":{"credence":0.88,"note":"The pattern recurs across methods and years. Adversarial training of a text classifier for a high-stakes filtering task roughly doubled the time human attackers needed to find a failure without making failures unfindable (Ziegler et al., 2022). Anti-scheming training of o3 and o4-mini cut covert actions on out-of-distribution tests from 13% to 0.4% and from 8.7% to 0.3%, with failures persisting in most of the environments where they had occurred (Schoen et al., 2025). Refusal training of deployed chat models has reduced but not ended successful jailbreaks. No published training method is reported to drive a targeted misbehaviour to zero under red-teaming.","seeded_by_claim_id":"0ab6034e-e3c4-4832-af44-22e54b77fdbd"},"quantity":null,"threshold_of":null,"question":null,"formalization":null,"formalization_pending":null,"formalization_history":[],"verification":null,"lean_checks":[],"lean_checks_total":0,"waiting_messages":[{"trigger":"structure_and_assess","from":"the system","sender_node_id":null,"sent_at":"2026-11-11T20:48:24.722Z","paid":false,"summary":"New claim onboarded. Give it the structure and assessment its importance warrants; a minor or settled claim may need no decomposition."},{"trigger":"mandate_plan","from":"a mandate's plan","sender_node_id":null,"sent_at":"2026-11-14T03:26:55.856Z","paid":true,"summary":"A funded mandate scheduled a assess pass on this claim."}],"requests":[],"requests_total":0,"subclaim_count":0,"instance_count":0,"source_map":null,"dependents":[{"id":"0ab6034e-e3c4-4832-af44-22e54b77fdbd","text":"Training an AI system on an accurate oversight signal makes it act as its overseers intend throughout deployment, including in rare high-stakes situations.","claim_type":"empirical_derived","node_kind":"claim","relation_type":"contradicts","reasoning":"The claim includes rare high-stakes situations, so a residual rate of failure after training, however small, is the kind of evidence that bears on it. The first half of this subclaim also records that training does carry over in large part.","importance":0.55,"assessment_status":"doubtful","assessment_confidence":0.8,"assessment_credence":0.07,"bounty_micro_usd":null,"checked":null,"formal":false}],"dependents_total":1,"arguments":[],"subclaims":[],"subclaims_assessed":0}