Meta has drawn a hard line in the AI arms race. The company’s Applied AI division now explicitly prohibits engineers from using Anthropic’s Claude Code and OpenAI’s Codex, two of the most popular AI-powered coding tools in the industry.
The restrictions, documented internally around June 29, 2026, are designed to prevent a specific and somewhat ironic problem: rival AI models accidentally teaching Meta’s own models their tricks.
The distillation problem
The concern is straightforward. When Meta engineers use Claude Code or Codex to write code, generate data, or run evaluations, those outputs carry the DNA of Anthropic’s and OpenAI’s proprietary models. If that output then feeds into Meta’s training pipelines, evaluation benchmarks, or post-training datasets, Meta’s models could absorb competitive capabilities they didn’t develop themselves.
That’s not just an intellectual property headache. It’s a potential breach of service agreements with both Anthropic and OpenAI, which typically prohibit using their outputs to train competing models.
