What did xAI use Claude models for?
Sources say xAI used Claude models for distillation and training
Multiple sources report that xAI used Claude models as part of its distillation and training workflows. The claim is that Claude was used specifically for tasks related to transferring capabilities into xAI systems (distillation) and supporting training, rather than being used only as an external chat interface.
The reporting also adds operational detail: sources say xAI’s process included using personal accounts and an intermediary service called Blackbox AI after xAI was cut off. In other words, after losing direct access to Anthropic’s models, xAI allegedly continued to obtain the model outputs and use them downstream.
This matters because model distillation is a core technique in modern AI development: a system can be trained to approximate the behavior of a larger or more capable model, often to reduce cost or improve efficiency. If xAI’s training pipeline relied on Claude outputs, that would suggest Claude played a meaningful role in shaping the behavior of xAI’s models.
It also highlights how “access” can become a compliance and policy issue when providers restrict usage. The allegation that xAI used alternative account setups and an intermediary tool after being cut off raises questions about adherence to those restrictions.
From an industry standpoint, the story is a reminder that competitive model ecosystems don’t operate in isolation—training supply chains can involve third-party models, and those relationships can change abruptly when access agreements shift.
Still, the broader technical specifics—such as exactly what distillation method was used, how prompts were structured, or what proportion of training data came from Claude—aren’t detailed in the provided summary. What is clear is the role attributed to Claude (distillation and training) and the reported workaround to maintain use after being cut off.