WHAT NEW IN X11.02 MON

AI collections from X · 22 collections · 64,204 posts

claudeIMAGE

Anthropic anti-distillation details: tens of thousands of accounts banned

Anthropic says anti-distillation cut extractable reasoning data by 90–99% and banned tens of thousands of accounts, often within 2 hours, at the cost of rejecting trillions of paid tokens and hitting legitimate developers.

tom_knock
@tom_knockkArchived 18 days agoOriginally posted 2026.10.13
Anthropic anti-distillation details: tens of thousands of accounts banned
tom_knock @tom_knockkOriginally posted 2026.10.13
Open original on X

Anthropic 抗蒸馏:官方说能抠走的 reasoning 数据,大概少了 90%–99%。 几条硬数字: 1. 平台上 distillers 能抽到的 reasoning tokens,降到原来的 1/10–1/100;有些天小到测不出 2. 摘要推理 vs 全转录:SWE-bench Verified 80.4% → 71.8% 3. Preserved thinking:8/31 后新 API 账号不能改 prior thinking 上下文;旧模型读不了 Fable 5.1 / Opus 5.5 的 reasoning 4. 两段式分类器 FPR <0.1%,强度对标生物武器过滤;Sonnet 5.5 / Opus 5·5.5 / Fable 5·5.1 全开 5. 9/24 起拦截请求也计费;数万账户被封,常 <2 小时 代价也写明了:拒掉万亿级 token 付费流量,合法开发者要改集成,误伤仍按拦截计费。

x.com/tom_knockk/status/2109857304502034440Archived 2026.10.14 10:19