Opus 4.6 represented target language internally before switching languages, with persistent Russian representations appearing before plausible textual cues

{
  "doc": "natural.md",
  "context": "NLAs revealed unverbalized language processing in Opus 4.6 that led to discovery of malformed SFT training data.",
  "norm_label": "Opus 4.6 represented target language internally before switching languages, with persistent Russian representations appearing before plausible textual cues",
  "graphify_id": "finding_language_switching_russian",
  "source_file": "natural.md",
  "imported_from": "/Users/antonborzov/Documents/Research.nosync/papers/extract_typed_out/natural/graph.json",
  "extracted_type": "finding",
  "source_location": "§Language Switching",
  "graphify_file_type": "finding"
}