# Anthropic rolls out stronger anti‑distillation safeguards

> Anthropic says its new defenses cut extracted reasoning data by 10‑100×, though some legit users must update integrations and may be charged for blocked requests.

Oossa · 2026-10-20 · https://oossa.com/en/anthropic-rolls-out-stronger-anti-distillation-safeguards

Anthropic announced that its latest anti‑distillation measures have dramatically lowered the amount of model reasoning data that attackers can harvest. The company reports a 10‑to‑100‑fold drop in extracted tokens after adding preserved thinking, a two‑stage classifier and stricter account limits. These changes affect Claude Opus 5.5, Claude Sonnet 5.5, Claude Fable 5.1 and similar models. Legitimate developers may need to tweak integrations, and Anthropic now charges for requests blocked by the new safeguards.

## The facts

- Distillation token extraction is now 10‑100 times lower than before.
- Preserved thinking blocks new API accounts created on or after August 31 2026 from editing prior context.

## Why it matters

Developers using Claude will see fewer disruptions from illicit copying attempts, but must adapt to new integration steps and potential charges for blocked calls.

## Sources & references

1. [Strengthening our defenses against distillation attacks](https://www.anthropic.com/news/distillation-defenses) – Anthropic

Last updated: 2026-10-20