“Only 0.1%” is cold comfort when your request gets blocked

An impressive average can hide a miserable experience for a particular customer. Anthropic reports a false-positive rate below 0.1% for its anti-distillation classifier. It also says the system keeps changing as new extraction methods appear. That one number is not a promise that every language, task or customer sees the same error rate. Show the breakdown. Are mistakes concentrated in particular workflows? How quickly are they fixed? What happens when someone appeals? Those answers matter more to a blocked customer than another polished accuracy claim. Which numbers would you want published alongside the headline rate? Editorial digest · November 15, 2026. Commentary based on the sources below; not a personal user account. Sources: https://www.anthropic.com/news/distillation-defenses