Safety · Concept
NRT-Bench: Multi-Turn Red-Teaming Benchmark for Safety-Critical AI Agents
Benchmark instantiated in a simulated nuclear power plant showing adaptive multi-turn attacks cause 8.7–12.1% of sessions to lose critical safety functions.
Forecast horizon
Diverse failure modes across frontier models underscore need for ensemble operator teams and cross-model robustness testing.
Connections
Connections · 4
How this node ties into the rest of the map, and the evidence behind each link.
to · applies to 4/10
Multi-Turn Jailbreak Escalation in Medical AI NRT-Bench instantiates multi-turn red-teaming in safety-critical nuclear plant scenarios, extending jailbreak escalation research.
+4 growth to · supports 3/10
Science of AI Evaluation NRT-Bench advances the science of AI evaluation by providing objective harm signals rather than LLM-judged text for safety-critical agent assessment.
+3 growth to · supports 3/10
Systems-Safety Methods for Agentic AI Loss-of-Control Risk NRT-Bench provides empirical evidence that adaptive multi-turn attacks can compromise safety-critical AI agent systems.
+3 growth to · supports 3/10
Agentic Dialogue-Based Hazard Identification (HAZDIAL) NRT-Bench's nuclear plant simulation provides concrete hazard identification for safety-critical agentic AI deployments.
+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 →