Anthropic says new measures like preserved thinking in Fable 5.1 and Opus 5.5 have reduced data harvesting through distillation attacks by about 90% to 99% (Anthropic)

Anthropic says new protections in Fable 5.1 and Opus 5.5 cut the effectiveness of distillation attacks by roughly 90% to 99%, materially reducing the risk that competitors or bad actors can harvest model behavior and replicate capabilities at scale. For CIOs, this elevates model security to a strategic buying criterion: organizations deploying frontier AI should evaluate anti-extraction controls, vendor safeguards, and how much proprietary value they expose through prompts, outputs, and integrations. IT leaders should treat AI governance as both a cybersecurity and IP-protection issue, not just a performance or cost decision.