Safety · Concept
GuardianBench: Same-Scene Instruction-Contrastive Embodied Risk Benchmark
Instruction-contrastive embodied safety benchmark; VLMs average only 24.1% pair accuracy on Safe/Unsafe pairs.
Forecast horizon
Connections
Connections · 3
How this node ties into the rest of the map, and the evidence behind each link.
from · applies to 4/10
UrbanGround: Real-Scale City Sandbox for MLLM Spatial Agency Both probe whether embodied/spatial MLLM agents remain reliable once perception must drive real-world action.
+4 growth to · tracked by 3/10
arXiv AI Safety Feed GuardianBench appears in the arXiv safety cluster on latent contextual risk in embodied AI.
+3 growth to · applies to 3/10
Risks of Mass AI Agent Interactions GuardianBench isolates latent contextual embodied hazards complementary to multi-agent interaction risks.
+3 growthSignal sources
Signal sources
Dated facts from primary sources in this direction.
US evaluation centre Jun 2025
In June 2025 the US AI Safety Institute was renamed the Center for AI Standards and Innovation (CAISI), pivoting toward security, standards and adversary-model assessment.
NIST → Frontier safeguards May 2025
Anthropic activated its ASL-3 deployment and security standard with Claude Opus 4 on 22 May 2025 — the first real-world trigger of a responsible-scaling tier, focused on blocking bio-weapon uplift.
Anthropic → Cross-border testing 2025
The International Network of AI Safety Institutes (launched Nov 2024) ran a third joint testing exercise focused on agentic AI systems across cyber and fraud strands.
European Commission — AI Office →