Meta bans rival AI code to prevent distillation attacks
meta restricted engineers from using claude code and codex to stop rival outputs from polluting its training data, signaling rising paranoia over model distillation.
meta has prohibited engineers from using anthropic's claude code and openai's codex after internal documents warned that rival outputs could leak into its training data. this restriction targets distillation attacks, where models are trained on the outputs of competitors to replicate their capabilities without authorization. meta is developing metacode to reduce reliance on expensive external tools and currently faces billions in internal ai costs. the move follows industry-wide friction, including anthropic accusing alibaba of massive distillation and musk admitting xai trained on openai outputs.