descriptive claim
Pre-specified AI capability thresholds have proven ambiguous in practice: model evaluation science is insufficient to determine whether a given model has definitively crossed a threshold, weakening the multilateral coordination case the thresholds were designed to produce.
desc_capability_thresholds_zone_of_ambiguity
confidence 0.90
Evidence (1)
supports (1)
- Anthropic's Responsible Scaling Policy: Version 3.0 primary_testimonyweight0.90
locator: Section: Assessing our theory of change
“We found pre-set capability levels to be far more ambiguous than we anticipated... The science of model evaluation isn't well-developed enough to provide dispositive answers.”