Meta has restricted its engineers from using Anthropic’s Claude Code and OpenAI’s Codex to prevent their outputs from entering Meta’s own training datasets. Internal documents reveal the company has temporarily halted specific work with these rival models to avoid distillation, which involves the unauthorized transfer of capabilities from one system to another. An internal memo warned that such leaks could cause serious escalations with partner companies if their proprietary model outputs were incorporated into Meta’s future training data.
This policy shift coincides with Meta’s development of its own coding assistant, MetaCode, and a strategy to reduce reliance on external tools due to rising operational costs. The company plans to spend billions on internal AI usage this year, while new rules bar employees from using AI outputs for test tasks or code analysis without human review. Distillation remains a contentious issue across the industry, evidenced by recent accusations against Alibaba and admissions from Elon Musk regarding xAI. OpenAI, Anthropic, and Google all explicitly ban using their model outputs to build competing systems in their terms of service.
- Meta’s policy bars engineers from using AI outputs for test tasks or code analysis
- Human review is still required for all internal AI usage
- The company plans to spend billions on internal AI use this year
