Internal Docs Show Meta Putting Limits on Claude and Codex, Fearing Distillation

As Meta Platforms tries to wean itself off expensive AI coding applications from Anthropic and OpenAI, it is confronting a difficult challenge: keeping employees from relying too much on those outside tools to build in-house replacements.

Internal guidelines reviewed by The Information show Meta is drawing strict limits on how engineers in its applied AI engineering division use Anthropic’s Claude Code and OpenAI’s Codex. One internal memo even instructed teams to pause some tasks that used those models over concerns that their output could seep into Meta’s training data—something the document warned could trigger “serious escalations with partner companies.”