Trends

Labs treat distillation of their models as a security threat

Adversarial distillation · Since Oct 2

Organizations

AnthropicFrontier Model ForumMoonshot AIOpenAI

Updates

  1. Oct 2New

    OpenAI shut down a reasoning-extraction campaign it tied to people associated with Moonshot AI, and Sonnet 5.5 shipped with anti-distillation classifiers.

    OpenAI disrupts a distillation campaignAnthropic's Claude Sonnet 5.5

Predictions

  1. Apr 2027

    Reasoning-extraction defenses and cross-lab information sharing become standard practice, and public attribution disputes among frontier labs grow more frequent. Oct 2