Meta just banned its own engineers from using Claude Code and Codex. Here's why your CISO should care.

Meta's Applied AI team can no longer use Anthropic's Claude Code or OpenAI's Codex without approval.

Not because they don't work. Because they work too well.

Internal documents obtained by The Information reveal Meta is terrified of "distillation" — the process where rival AI model outputs leak into your training data. An internal memo warned this could trigger "serious escalations with partner companies."

Read that again. A $1.5 trillion company just admitted that its own engineers feeding code into competitor AI tools is a contractual risk, not just a competitive one.

Here's what's actually happening. When your engineers use Claude Code to debug a training script, chunks of that codebase travel to Anthropic's servers. That's normal. But when the codebase is your proprietary AI training infrastructure, you've just given your competitor a free sample of your secret sauce.

Meta is now building its own coding assistant, MetaCode. The restriction targets engineers working directly on model building — not the entire engineering org. But the signal is clear: frontier AI labs are treating rival model outputs as supply chain contamination.

This is the new enterprise reality. Your developers are already using Claude, Copilot, and Codex. Every line of code they feed these tools is a data point. If you're building proprietary models, you need to audit every AI coding tool your team touches — and ask what happens to the data after it leaves your network.

The distillation war just went from theoretical to boardroom. Audit your AI toolchain today.

SOURCE: https://www.theinformation.com/articles/internal-docs-show-meta-putting-limits-claude-codex-fearing-distillation
VERIFIED: The Information, Crypto Briefing, ZeroHedge, Business Today
SIGNAL: Meta's restriction proves that AI coding tools are now a competitive attack surface — not just a productivity play. Every enterprise building proprietary models faces the same exposure.