Meta Locks Claude and Codex to Protect Its AI Data

Meta Locks Down Claude and Codex to Protect Its AI Data

Meta is limiting the use of Claude Code from Anthropic and Codex from OpenAI to prevent the results from these AI tools from ending up in its own training data. According to internal documents obtained by The Information, Meta has even temporarily suspended certain work with these models.

The company is concerned about distillation—the unauthorized transfer of capabilities between competing AI models. An internal memo warned of potential serious escalations with partner companies if the results from their models were to leak into Meta's training data.

Meta is currently developing its own coding assistant, MetaCode, and aims to reduce its reliance on external tools, partly due to rising costs. According to an internal memo, the company is on track to spend billions of dollars on internal AI usage this year alone. The company's policy prohibits engineers from using AI results to create test tasks or for code analysis. Human review is always required.

Distillation is causing tensions in the industry. Anthropic recently accused Alibaba of conducting the largest known distillation attack to date, and Elon Musk had to admit in April that xAI had partially distilled OpenAI's models. The terms of service for OpenAI, Anthropic, and Google explicitly prohibit the use of model results to build competing systems. Meta stated that it has clear rules for the responsible use of AI tools.