Safety · Concept
Beneficial Trait RL for Broad Alignment Generalization
Post-training domain (helpfulness vs. coding) differentially degrades mid-trained compassion values, with helpfulness training causing 35.7% vs. 65.2% retention on animal harm benchmark.
Forecast horizon
Helpfulness training degrades general moral reasoning by 25.5 pp, raising concerns about standard RLHF pipelines.
Connections
Connections · 2
How this node ties into the rest of the map, and the evidence behind each link.
to · constrains 4/10
Emergent Misalignment from Narrow Fine-Tuning Beneficial trait RL training generalizes alignment to out-of-distribution benchmarks, potentially countering emergent misalignment from narrow fine-tuning.
+4 growth to · supports 2/10
RLHF Shallow Alignment and Partisan Structure Preservation Helpfulness post-training degrades mid-trained compassion values and general moral reasoning, consistent with RLHF shallow alignment concerns.
+2 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 →