Beijing's approach to xAI is one lawsuit against its own users. Grok's parent company sued a group of its users this month over CSAM Grok generated, because a March 2026 Texas ruling stripped Section 230 protection from AI outputs the platform itself produces (the law that shields platforms from liability for what *users* post never covered content the model generates unprompted). xAI's fix wasn't a filter update. It was suing the people who typed the prompts, converting a product-safety failure into a plaintiff's case against its own customer base.

The maneuver matters beyond Grok because it is a template. Any lab whose output triggers liability now has a playbook: argue the user's prompt, not the model's weights, produced the harm, and let discovery sort out whose fingerprints are really on the image. Apple's July 9 trade-secrets suit against OpenAI runs the same discovery machinery in the other direction, subpoenas into what OpenAI's own systems generated and how. Both cases will produce the first hard case law on how much of an AI model's output a company can disclaim as someone else's doing, and that ruling, not any lab's usage policy, sets the liability floor every deployment from Sydney to Singapore prices against next.