Meta restricts engineers from using Claude Code and Codex, citing AI model distillation risk
Category: AI Governance
What it is
Meta has reportedly barred engineers in its Applied AI division from using
Anthropic’s Claude Code and OpenAI’s Codex without approval, per The
Information. The concern is model distillation – outputs from rival tools potentially
teaching Meta’s in-house coding model, MetaCode, capabilities it didn’t develop
itself, risking IP exposure since both vendors’ terms of service prohibit training
competing models on their outputs. Usage had reportedly reached 60 trillion
tokens in 30 days before the restriction. Microsoft has also pulled back Claude
Code licenses, citing token consumption.
Why it Matters for Enterprises
AI-to-AI interaction is now a compliance issue, not just human data handling.
Enterprises building in-house AI tools should audit whether staff usage of third
party coding assistants could contaminate training data or breach vendor terms.
