Meta restricts engineers' use of Claude Code and Codex over distillation risk

Meta has imposed strict limits on how engineers in its applied AI division can use two coding tools: Anthropic's Claude Code and OpenAI's Codex. The Information reported the restrictions, citing sources.

Why Meta wants its own coding tools

Meta wants to build its own tools for automating code writing. At the same time, it is reducing its engineers' dependence on the competing products they use today. While Meta works on in-house alternatives to Claude Code and Codex, leadership is asking engineers in the applied AI division to be more careful with external tools.

Claude Code is Anthropic's agent that works autonomously with a codebase from the terminal, powered by the Claude family of models. Codex is OpenAI's counterpart: it integrates with GPT models and handles similar development automation tasks. Over the past few years, both have become some of the most prominent AI agents on the market for writing and refactoring code.

What the distillation risk is

Meta's official reason for the restrictions is the risk of inadvertent distillation. The industry uses this term for cases where heavy use of a third-party AI tool can unintentionally hand a competitor signals about a company's internal processes, code patterns, or approaches. That, in turn, could help Anthropic or OpenAI improve their own models using Meta's work.

• The restrictions apply to engineers in Meta's applied AI division
• They cover the two most prominent AI coding agents on the market: Claude Code (Anthropic) and Codex (OpenAI)
• The official reason is the risk of inadvertent distillation
• The Information first reported the news, citing sources
• In parallel, Meta is building its own coding tools to reduce its dependence on competitors' products

Who else is in this race

The situation shows how the biggest AI labs act as both suppliers and competitors to one another. Anthropic and OpenAI sell access to their development tools to companies like Meta, which competes with them in foundation models and plans to bring its own AI coding products to market. This kind of mutual dependence is common across the industry. Nearly every major tech company uses outside AI services internally while building competing products. That makes protecting internal data and work especially sensitive in the internal teams that build a company's most valuable products.

Why this matters now

The race to build the best AI tool for developers runs alongside the race to build the best large language model, and that is no coincidence. Engineers working on the most complex and proprietary parts of a codebase give an external AI agent detailed technical tasks every day, show it pieces of internal architecture, and get generated code back. For a company of Meta's scale, which is investing billions in its own AI models while remaining a customer of its direct competitors, this daily routine becomes an information security issue on par with protecting source code or user data.

What it means

Meta's restrictions show that big tech companies are starting to treat third-party AI development tools not just as a way to boost engineer productivity, but also as a potential channel for leaking competitive advantages. As AI coding agents become a core part of developers' daily work, the question of who ultimately gets data on how a company writes its code is growing in strategic importance. That is especially true for a player of Meta's scale, which is both a customer and a direct competitor of Anthropic and OpenAI in the AI market.