safety security
The Gate Counts the Cost of Being Wrong
In simulation, routing tool calls by consequence, reversibility, ambiguity and evidence freshness cut harmful-side-effect trajectories from 14.75 to 7.02 percent at equal compute cost.
Summary
In simulation, routing tool calls by consequence, reversibility, ambiguity and evidence freshness cut harmful-side-effect trajectories from 14.75 to 7.02 percent at equal compute cost.
Runtime safeguards for tool-using agents often ask only how uncertain a proposed action looks. CASEL instead assigns each call to direct execution, verification, sandboxing, clarification, approval or abstention by combining calibrated failure probability with state-dependent consequence, reversibility, ambiguity and evidence freshness. Across 20 simulated seeds with 120,000 test actions each, its allocation policy reduced harmful-side-effect trajectories from 14.75 percent under an uncertainty-ranked gate to 7.02 percent and lowered normalized harm from 0.184 to 0.123 at the same compute cost. A simpler probability-times-consequence ranker reached similar harm but sent 45 percent more actions to human approval. These results compare policies inside a controlled simulator whose intervention effectiveness is assumed, not measured; they do not establish deployment performance.
Why it matters
In simulation, routing tool calls by consequence, reversibility, ambiguity and evidence freshness cut harmful-side-effect trajectories from 14.75 to 7.02 percent at equal compute cost.
Limits and context
- Runtime safeguards for tool-using agents often ask only how uncertain a proposed action looks.
- These results compare policies inside a controlled simulator whose intervention effectiveness is assumed, not measured; they do not establish deployment performance.
Key claims
In simulation, routing tool calls by consequence, reversibility, ambiguity and evidence freshness cut harmful-side-effect trajectories from 14.75 to 7.02 percent at equal compute cost.
Qualification: Runtime safeguards for tool-using agents often ask only how uncertain a proposed action looks.
Evidence: source-2026-10-28-001
Sources
- arXiv preprint 2610.36570, original v1arXiv · primary research
Corrections
No corrections have been recorded for this story.