Claude Fable 5 secretly throttled AI researchers, and the internet went wild

ZDNET covers the controversy around Anthropic's Fable 5, which initially employed invisible downgrades to a less-capable model (Opus) for flagged frontier tasks; researchers assumed they were getting Fable results, triggering backlash over transparency, defensive research limitations, and 30-day data retention policies. The article details expert reactions, Anthropic's apology and move to make downgrades visible, and the debate over whether such safeguards help or hinder defenders and whether motivated attackers will circumvent them.