Anthropic's preserved thinking blocks account-switching distillation attacks
ClaudeDevs · x · 2026-09-29
Anthropic is extending preserved thinking to Claude Sonnet 5.5 to curb distillation attacks via account-switching. Thinking blocks are now tied to the org that produced them: if you switch accounts mid-session, Claude rereads the session and regenerates its thinking. The API also verifies each block's signature—whether the current model can read it and whether the prefix (system prompt, tools, prior messages) is unchanged—rejecting or dropping blocks on mismatch. Prefix enforcement is default for accounts created after Aug 31, 2026; Anthropic recommends append-only integrations for everyone.
Related event: Anthropic Launches Claude Sonnet 5.5: Over 30% Faster, Up to 30% Cheaper(58 posts)→
More from Models
- 'Sandbox escape attempts' way down but nobody knows why — alignment researchers push embedded evals — peterwildeford · 2026-11-13
- Jason Calacanis threatens to ditch $20k/yr Claude as $40/mo Grok Bot outperforms it — chrisfleck · 2026-11-13
- Dan Shipper's weekend plan: let Claude Fable 5.2 build a custom Supernote OS — danshipper · 2026-11-13
- Alibaba teases Qwen4 with stronger coding and agentic capabilities at Nov 17 Hong Kong summit — teortaxesTex · 2026-11-13
- Baidu opens ERNIE 5.5 Flash Preview via OpenRouter — Baidu_Inc · 2026-11-13
- Hybrid Decision Model learns when to reason: 72.5% lower latency at higher accuracy — hbouammar · 2026-11-13