In order to prevent the accidental leakage of its proprietary technologies, tech giant Meta has begun to strictly regulate its internal AI development tools. According to insiders, Meta has already established clear guidelines internally, restricting engineers in its AI engineering department from using Anthropic's Claude Code and OpenAI's Codex. Notably, this decision has nothing to do with corporate costs or employee productivity, but rather stems from concerns about core technology security.

Strictly Preventing "Model Distillation" from Competitors

This internal document, which was issued as early as May this year and has been in effect since, reveals Meta's core concerns. Meta is extremely worried that engineers might unintentionally trigger so-called "distillation behavior" when using these third-party tools to assist in programming, which involves using competitors' model outputs to reverse-train its own AI models. This operation is not only sensitive in the tech industry but also directly violates the user service agreements of Claude and Codex.

To avoid potential legal risks, Meta has currently asked relevant technical teams to immediately halt high-risk tasks that depend on the aforementioned models. The company warned internally that if competitors' AI core outputs seep into Meta's own training datasets, it could lead to serious legal disputes with partner companies, even escalating the situation.

Policy for Responsible Development

In response to this restriction, a Meta spokesperson publicly commented. The spokesperson emphasized that the company has always had very clear and strict policies to regulate how teams use external AI tools, aiming to ensure that engineers can focus on high-impact core R&D work in a responsible manner.