한국어EN
Safety, Alignment & InterpretabilityAnthropic FellowsAnthropic

Model Spec Midtraining

Made by AIWritten by AI (Claude) from the source below. Check anything important against the source.How it's made

Model Spec Midtraining

Why it matters

Putting a norms document into midtraining, before alignment fine-tuning, improves generalization

Source — Primary sources first; secondary reports are labelled “Press”

Officialalignment.anthropic.comhttps://alignment.anthropic.com/2026/msm/

Added in Week 35, 2026 · See that week

On the map — ⑩ Safety, Alignment line

  1. Previous stopClaude Opus 4.6
  2. This recordModel Spec Midtraining
  3. Next stopShieldstral
See it on the map

Same week

Iwase et al. publish Reliable Chain-of-Thought via Prefix ConsistencyPaperTeaching Claude WhyOfficialStop Comparing LLM Agents Without Disclosing the HarnessPaperGPT-Realtime-2 / -Translate / -WhisperOfficial

Nearby on the same axis

Teaching Claude WhyOfficialAnthropic proposes an Advanced AI Framework for policyOfficialChina restricts OpenClaw use at state-owned enterprises and government agenciesPressAgentic Misalignment in Summer 2026 (14 models)Official