Meta bars Claude Code and Codex from training pipelines over distillation risk
Meta has restricted internal use of Claude Code and OpenAI Codex — blocking outputs from test-task generation and code analysis — to prevent rival AI capabilities from entering its own training data via distillation, with internal memos warning of 'serious escalations' if model outputs leaked. The restriction is live engineering policy, not a draft guideline: ToS from OpenAI, Anthropic, and Google already ban using model outputs to train competing systems, and Meta's clampdown confirms that boundary is now enforced at the operational level inside the largest AI shops.
Source: the-decoder.com ↗
Meta is limiting how its engineers use Anthropic's Claude Code and OpenAI's Codex to prevent outputs from these AI tools from ending up in its own training data.
Why this matters
- → Distillation enforcement moves from ToS to actual operational policy inside major AI labs
- → Meta's billions in annual AI spend now constrained by intellectual property boundaries
- → Industry friction escalates as competitors guard model capabilities at source