Safety · Concept
Agreement Is Not Alignment: Divergent Moral Grounds
Shows high label agreement between humans and LLMs can mask systematic divergence in moral rationales and principles.
Forecast horizon
Connections
Connections · 4
How this node ties into the rest of the map, and the evidence behind each link.
from · frames 4/10
Theory of Value in AI Alignment Research Both critique preference/label agreement as an inadequate stand-in for genuine value alignment.
+4 growth to · supports 4/10
Theory of Value in AI Alignment Research Rationale-level moral divergence shows preference/label agreement is an inadequate theory of value for alignment.
+4 growth to · informs 3/10
Cognitive Alignment: AI Mirroring Human Reasoning Divergent moral grounds despite label agreement motivate calls for cognitively aligned rationales users can understand.
+3 growth from · requires 3/10
Cognitive Alignment: AI Mirroring Human Reasoning Mirroring human reasoning and communicating rationales is needed because matched final labels do not imply shared moral grounds.
+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 →