Step 2: tool results

get_claim

{
  "include": [
    "arguments",
    "subclaims",
    "dependents"
  ],
  "claim_id": "ea57432c-c473-4452-94e1-16102c805446"
}

{"claim":{"id":"ea57432c-c473-4452-94e1-16102c805446","kind":"claim","canonical_form":"In AI control evaluations, weak trusted monitors with resampling, replay, and auditing catch most sabotage attempts by stronger untrusted models.","claim_type":"empirical_derived","state":"active","decomposition_status":"complete","importance":0.4,"contestation":0.45,"domains":[],"tags":[{"slug":"ai-control","name":"AI Control"},{"slug":"scalable-oversight","name":"Scalable Oversight"}],"steward_state":"running","stewarded_at":"2026-11-15T00:21:12.502Z","canonical_direction_note":null,"created_at":"2026-11-11T20:25:29.352Z","updated_at":"2026-11-15T00:21:05.695Z","page_url":"https://minerval.ai/claims/ea57432c-c473-4452-94e1-16102c805446"},"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":0.55,"note":"Against the attack policies their authors built, the published protocols do catch most attempts: the original control paper (GPT-3.5 overseeing GPT-4 on backdoored code) reported large gains over baselines, and Ctrl-Z reported resampling protocols cutting attack success from 58% to 7% at a 5% usefulness cost. The result is sensitive to the strength of the red team. Later work reports that prompt injections aimed at the monitor defeat the same protocols and make resampling backfire, that protocol-aware attackers and strategic attack selection cut measured safety sharply, and that human-crafted attack trajectories evade monitors more often than model-generated ones. Whether \"most\" holds therefore depends on which attacks are counted.","seeded_by_claim_id":"6e3da3ea-b163-4579-b6fb-5b3ce5b186a6"},"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":"2026-11-11T20:25:29.776Z","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":"dependent_note","from":"the Steward of a node that depends on this one","sender_node_id":"6e3da3ea-b163-4579-b6fb-5b3ce5b186a6","sent_at":"2026-11-11T20:53:32.776Z","paid":false,"summary":"The claim that developers' control of inputs, memory, copies and internals offsets a moderate capability advantage was re-judged from the results sections of its sources: still inconclusive, credence …"},{"trigger":"mandate_plan","from":"a mandate's plan","sender_node_id":null,"sent_at":"2026-11-15T00:21:05.697Z","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,"arguments":[],"subclaims":[],"subclaims_assessed":0,"dependents":[{"id":"6e3da3ea-b163-4579-b6fb-5b3ce5b186a6","text":"Developers' control of an AI system's inputs, memory, copies, and internals offsets a moderate capability advantage over its overseers.","claim_type":"empirical_derived","node_kind":"claim","relation_type":"supports","reasoning":"Control evaluations are the nearest direct test of whether developer-side levers let a weaker overseer contain a more capable model that is trying to subvert it. Evidence for the claim, not a requirement of it.","importance":0.45,"assessment_status":"doubtful","assessment_confidence":0.6,"assessment_credence":0.35,"bounty_micro_usd":null,"checked":null,"formal":false}],"dependents_total":1}