Safety · Concept
MedFailBench: Medical AI Safety Boundary Benchmark
Clinician-built open benchmark labeling medical AI failures by severity and safety-gate type rather than only answer correctness.
Forecast horizon
Public v0.2.1 release enables comparative safety-boundary inspection without patient data.
Connections
Connections · 4
How this node ties into the rest of the map, and the evidence behind each link.
to · applies to 4/10
Science of AI Evaluation MedFailBench reframes medical evaluation around which safety boundary failed rather than only knowledge accuracy.
+4 growth to · constrains 4/10
Cura 1T: Specialized Agentic Healthcare Model MedFailBench inspects which medical safety boundaries fail, providing an evaluation foil for specialized healthcare agents/models.
+4 growth to · applies to 4/10
Cura 1T: Specialized Agentic Healthcare Model MedFailBench safety-boundary inspection complements specialized agentic healthcare model evaluation beyond accuracy.
+4 growth from · publishes 3/10
arXiv AI Safety Feed arXiv safety cluster includes the MedFailBench medical safety boundary benchmark.
+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 →