Evidence · DiggingBeagle record
CASEL: Consequence-Aware Selective Execution for Tool-Using Agents
Simulation of consequence-aware tool-call intervention over 20 seeds and 120,000 actions per seed. The authors disclose assumed intervention effectiveness, so harm reductions are comparative model outcomes, not deployed risk reductions.
- Source role
- primary disclosure
Evidence record
Simulation of consequence-aware tool-call intervention over 20 seeds and 120,000 actions per seed. The authors disclose assumed intervention effectiveness, so harm reductions are comparative model outcomes, not deployed risk reductions.
Claim-level citations (3)
- supportsConsequence-aware tool gating reduced simulated harmful side-effect trajectories: The paper simulates 20 seeds with 120,000 test actions per seed and reports harmful trajectories falling from 14.75% to 7.02% versus an uncertainty-ranked gate.
Abstract; controlled intervention-menu comparison
- supportsConsequence-aware tool gating reduced simulated harmful side-effect trajectories: The authors report normalized mean harm moving from 0.184 to 0.123 under matched simulated action budgets.
Abstract; consequence and cost measurements
- supportsConsequence-aware tool gating reduced simulated harmful side-effect trajectories: A gate using estimated failure probability multiplied by consequence reached 5.82% harmful trajectories in the same simulation, but routed 45% more actions to human approval than CASEL.
Abstract, cost-matched risk-product comparator