Benchmark RadarAI BENCHMARK PROFILE
OSGuard
OSGuard is a dual-granularity benchmark suite for evaluating safety in computer-use agents, with an action-level benchmark for local guardrail decisions and a risk-augmented execution suite for end-to-end evaluation.
- Released
- 2026-06-13
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
Task success alone misses unsafe shortcuts in computer-use agents. OSGuard's dual-granularity design distinguishes local recognition of unsafe actions from full-task safety, providing a more precise diagnosis of guardrail capabilities.
Motivation
Computer-use agents are increasingly evaluated by whether they complete realistic desktop and web tasks.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.