Meta Reportedly Limits Use of Claude Code, Codex Over AI Distillation Concerns

Meta has reportedly restricted engineers in its Applied AI division from freely using AI coding assistants developed by rivals Anthropic and OpenAI, amid concerns that interactions with competing models could inadvertently contribute to the replication of proprietary AI capabilities.

Engineers in Meta's Applied AI team are now required to obtain approval before using Anthropic's Claude Code or OpenAI's Codex. The reported policy is aimed at reducing the risk of Meta's internal AI systems interacting with competing models in ways that could unintentionally expose or reproduce proprietary knowledge.

Also Read: India's UPI Goes Live in Greece, Expands European Payment Footprint

Model Distillation:

According to reports, the concern centres on model distillation, a technique that enables a smaller AI model to learn by replicating the behaviour of a larger, more advanced model. While distillation is widely used to create faster and more cost-efficient AI systems, it has also become a growing source of concern for leading AI developers.

Companies building frontier AI models increasingly argue that extensive interactions with proprietary systems could allow competitors to recreate key capabilities without direct access to the underlying models, raising legal, commercial and ethical questions around intellectual property.

If confirmed, the move would represent one of the first publicly reported instances of a major AI company restricting employee access to competing AI coding tools specifically over concerns related to model distillation.

Also Read: WhatsApp Begins Rolling Out Usernames, Adds New Privacy Controls

Enterprise AI:

Meta has not publicly confirmed the reported policy. It also remains unclear whether the restrictions apply only to the Applied AI division or extend to other engineering teams within the company.

The reported measures also underscore broader concerns around enterprise AI adoption. Although Anthropic and OpenAI offer enterprise versions of their AI products with enhanced security and privacy controls, companies continue to debate whether interactions with external AI systems could expose sensitive internal code, confidential information or intellectual property.

Meta is not alone in limiting employee use of external AI tools. Apple, Samsung and Amazon have previously imposed restrictions on services such as ChatGPT and GitHub Copilot over concerns that proprietary code and confidential business information could be inadvertently shared with third-party AI platforms.

The development highlights the increasing emphasis AI companies are placing on safeguarding proprietary models as competition intensifies and concerns over AI training methods, intellectual property and model replication continue to grow.