Current outcome
Sonnet 5.5's anti-reasoning-extraction protections have begun showing effects: Anthropic says extraction has dropped sharply, tens of thousands of distillation-related accounts have been banned, and charging has resumed for some blocked distillation-related requests.
Progress timeline
4 material updates- #01
Claude Sonnet 5 - Anthropic
Anthropic announced Claude Sonnet 5, a new iteration of its mid-tier Sonnet model family.
Source evidence: Anthropic News
- #02
Introducing Claude Sonnet 5.5 - Anthropic
Anthropic announced the release of Claude Sonnet 5.5, the latest iteration of its mid-tier frontier model family.
Source evidence: Anthropic News · Anthropic News · TechCrunch AI
- #03
Sonnet 5.5成首个防「推理抽取」的Sonnet：内部思考开始绑定账号
Anthropic added anti-reasoning extraction safeguards to Sonnet 5.5: generated thinking blocks are bound to the originating account or organization, and a classifier detects large-scale harvesting of reasoning for distillation.
State after update: Sonnet 5.5 has been released and is the first Sonnet model with anti-reasoning extraction account binding and a classifier.
Source evidence: theblockbeats
- #04
Anthropic强化蒸馏攻击防御，已封禁数万个相关账号
Anthropic disclosed initial results of its Sonnet 5.5 anti-distillation measures: extraction of Claude reasoning tokens by distillation attackers fell roughly 90%–99%, tens of thousands of related accounts were banned, and charging resumed since September 24 for blocked distillation-related requests.
State after update: Sonnet 5.5's anti-reasoning-extraction protections have begun showing effects: Anthropic says extraction has dropped sharply, tens of thousands of distillation-related accounts have been banned, and charging has resumed for some blocked distillation-related requests.
Source evidence: theblockbeats