Meta has instructed engineers in its Applied AI division to limit or restrict their use of Anthropic’s Claude Code and OpenAI’s Codex, according to internal documents reviewed by The Information and reported by outlets including The Next Web and ZeroHedge. The guidance targets concerns about “inadvertent distillation,” where high-quality outputs from rival AI systems could unintentionally be incorporated into Meta’s own development pipeline. The issue is framed as a risk that code suggestions, architectural guidance, debugging help, or other generated artifacts from Claude Code or Codex could “contaminate” data used to train or improve Meta’s models, including its Llama family.
The restrictions are described as active by late June and referenced as originating at least as far back as May, based on the internal documents. The guidance reportedly applies more narrowly to teams working directly on model building and applied AI initiatives rather than to the entire engineering organization. Meta has not publicly confirmed the directive, and the reports say it has not commented. The overall aim described across sources is to reduce the chance that competitor model behavior becomes transferred into Meta’s systems through routine use.
