Safety · Trend
Capability evaluations
Pre-deployment tests for cyber, bio and autonomy risks are becoming a condition of release.
Forecast horizon
Evaluations get standardized and audited; the science of 'what a test proves' has to catch up.
Connections
Connections · 6
How this node ties into the rest of the map, and the evidence behind each link.
from · institutionalizes 7/10
AI safety institutions Institutes turn evaluation into standing public capacity.
+3 growth from · tracked by 6/10
AI agents Autonomy raises the stakes for pre-deployment evaluation.
+3 growth to · informs 6/10
Frontier safety frameworks Evaluation results set the thresholds in safety frameworks.
+2 growth to · tracked by 6/10
AI agents Agentic autonomy is the hardest thing to evaluate.
+3 growth from · standardizes 5/10
Intl safety-institute network The network pushes shared evaluation methods.
+3 growth from · publishes 5/10
UK AI Safety Institute The UK institute publishes open evaluations.
+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 →