From the blogAll posts →
- Claude’s Watermark Is Not EncryptionAnthropic announced on August 11 that Claude models launched from August 2, 2026 onward will embed an invisible watermark in generated text, worldwide, as part of implementing its…
WritingAll writing →
Essays, notes, spines, and readings on measurement, meaning, and human oversight of AI systems.
Research threadsAbout the research →
01Evaluation and measurementground truth · instrumentation · validity
02Information and meaningsemantics · embeddings · context
03Affective computingemotion · judgment · machine behavior
04Governancestandards · taxonomies · policy
05Human-AI collaborationoversight · teaming · supervision
