Benchmark RadarAI BENCHMARK PROFILE
DeCompBench
DeCompBench evaluates agent safety against decomposition attacks, where harmful tasks are broken into benign subtasks; it measures refusal rates and objective fulfillment on decomposed variants.
- Released
- 2026-06-12
- Readiness
- Inspectable
- Primary field
- General AI
Why it matters
Addresses a security gap not covered by existing agent safety benchmarks, critical for preventing adversarial misuse in deployed agents.
Motivation
LLM-based Agents are becoming increasingly capable and widely deployed, creating growing incentives for adversarial misuse in the real-world.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.