Meta restricts employee access to Anthropic Claude Code, OpenAI’s Codex over distillation risks

Meta has reportedly restricted employees from accessing AI tools from rivals over the risk of unintentionally extracting proprietary knowledge. According to The Information report, Engineers in Meta's Applied AI team can no longer use Anthropic’s Claude Code or OpenAI’s Codex without approval. 

It is suggested that Meta's internal AI systems interact aggressively with these rival AI models, which could lead to unintentional extraction of knowledge from Anthropic's or OpenAI's proprietary models. This could create potential legal or ethical concerns for Meta as distillation is viewed as copying the behaviour of another company's AI model without authorisation.

Must read: Google caps Meta's Gemini AI usage over compute limits

This comes as the earliest publicly reported cases of a leading AI company restricting employees from using competitors' AI coding tools specifically because of concerns over copying or learning proprietary capabilities from competitors' models. 

On the other hand, Meta has not confirmed these claims as of now. In addition, it is also not confirmed whether the restriction is across Meta's engineering team or some specific teams working on AI model development.

In addition, the restriction also raises concerns over the gap in current enterprise AI offerings. While both Anthropic and OpenAI already sell enterprise versions of their AI products, but it may still not fully address concerns about sensitive internal data being exposed in ways that could affect AI model training or intellectual property.

Must read: WhatsApp Username rules explained: Allowed characters, length, and restrictions you need to know

It should also be noted that Meta is not alone, as other tech giants including Apple, Samsung, and Amazon have previously restricted employees' use of OpenAI’s ChatGPT or corporate GitHub Copilot versions due to fears of leaking proprietary code or trade secrets into external LLM training pools. Therefore, while AI tools are helping companies be more efficient, there’s also a risk of sensitive data and proprietary information.