Anthropic model card cites synthetic data

This is less a scandal than a transparency moment: synthetic data is standard now, but the wording is easy to overread as a confession of competitor-model training.

• –Anthropic’s current system cards explicitly mention synthetic data alongside deduplication and classification in the training pipeline.
• –That wording does not by itself prove they trained on GPT or Gemini outputs; “other models” can also mean internal or partner systems.
• –The important issue for developers is provenance: once synthetic data enters the stack, model lineage gets fuzzier and auditability gets harder.
• –The Reddit reaction shows how sensitive training-data disclosures remain, especially when companies criticize rivals for distillation and data scraping.