Meta Internally Restricts Engineers' Use of Claude Code and Codex to Avoid Distillation Trap and Reduce Costs

Meta has recently enforced an internal policy that restricts its engineers from using external AI coding models, including Claude Code and Codex. The restriction aims to prevent the outputs of these models from leaking into the training and evaluation data pipeline of Meta's self-developed AI coding assistant, MetaCode. By doing so, Meta seeks to avoid falling into the "distillation trap" — a scenario where it unintentionally copies the capabilities of competing models. Additionally, limiting reliance on external AI services helps reduce the significant usage costs associated with such tools. As a result, Meta developers will be expected to rely primarily on MetaCode for their coding assistance needs. This policy underscores the growing emphasis big tech companies place on data purity and cost efficiency as they race to develop proprietary AI solutions.