Meta has implemented internal restrictions preventing its engineers from using Anthropic's Claude Code tool and other advanced coding assistants, according to internal documents obtained by The Information and reported by Crypto Briefing. The limits appear designed to protect Meta's proprietary AI training data and intellectual property from potential exposure through Anthropic's systems. While specific details about the scope and timing of the restrictions remain limited, the move signals growing concern among major tech companies about how their internal work and codebase could be leveraged by competing AI systems during training phases. Meta did not immediately respond to requests for comment on the policy.
The restrictions reflect broader industry anxiety around model distillation—the practice of extracting knowledge from one AI system to train a competing model. By limiting engineer access to Claude Code, Meta aims to prevent its internal code, architectural decisions, and proprietary development practices from potentially being captured and used to improve rival systems. This concern has become increasingly material as companies compete to develop advanced AI capabilities. Anthropic has not publicly commented on Meta's restrictions, though the move underscores the stakes around data governance in competitive AI development environments.
The incident occurs amid other significant Anthropic developments, including the company's recent U.S. regulatory clearance to restore its most powerful models for select companies and a new partnership with California's government to deploy Claude across state operations. These parallel moves illustrate Anthropic's expanding institutional footprint even as individual corporations take defensive measures. The tension between Anthropic's growing enterprise and government adoption and corporate guardrails around its tools suggests the market for advanced AI capabilities remains defined by deep competitive and security concerns—not yet by widespread trust in how coding assistants handle sensitive organizational data.