Anthropic and pretraining
Claude models
Pretraining data cutoff dates
See the fog app for an interactive view of models’ epistemic horizons.
(table)
Methods
- Decontamination
Methods
- Modular Pretraining (GRAM)
Data filtering
- Classifier-based data filtering1
- Canary strings
External links
Discussion
- Anthropic’s focus on hyperstition (Simon Lermen, May 11, 2026)
Footnotes
-
Enhancing Model Safety through Pretraining Data Filtering (Aug 19, 2025) ↩