Anthropic cuts distillation attack exposure 65x with new Claude safeguards
CurieuxExplorer · x · 2026-10-13
Anthropic says it has significantly reduced the effectiveness of distillation attacks against Claude. Official data shows daily reasoning-token exposure to successful distillation requests averaged 8,800x a quiet day before safeguards (peaking at 56,000x), then dropped to an average of 135x after full rollout on Sept 19 — roughly a 65x reduction.
The thread highlights the distillation paradox: the same technique that makes models smaller and cheaper can extract proprietary capabilities without authorization, raising hard questions about where to draw the line between protecting and enabling innovation.
More from Models
- Apollo: Gemini 4 Argon greeted evaluators in its chain-of-thought during evals — OwariDa · 2026-10-14
- More thinking, worse scores: refusals tank closed-model evals at max reasoning effort — _lewtun · 2026-10-14
- Google Is Definitely Back With Argon, Says Investor, May Soon Close Gap With OpenAI and Anthropic — bindureddy · 2026-10-14
- Users say OpenAI's newest frontier models are worse and pricier than GPT-5.3-codex — this_nice_demon · 2026-10-14
- Argos High keeps saying "I'm just a language model" inside Gems — Necessary-Ad-6088 · 2026-10-14
- Paying for Both: This User Says ChatGPT Loses to Claude at Everything but Image Generation — Pretend_Painter_7099 · 2026-10-14