Cursor probed whether employee Grok use exposed client data to SpaceX

Cursor, the AI coding platform, investigated whether its own employees may have sent confidential client data into SpaceX’s AI systems while testing Grok. The timing is awkward. The probe reportedly started in late July 2026, just weeks before SpaceX finalized its $60 billion acquisition of Cursor on August 14, 2026. What set off the investigation The probe began after engineers raised concerns on Slack. They were worried about how confidential client data was being handled during pre-launch testing of the Grok Bot. At the center of those warnings was an internal tool called Sand. Employees flagged the risk that Sand might have inadvertently routed sensitive information to SpaceX’s AI division for use in model training. Cursor’s legal team took the warnings seriously. A company lawyer cautioned that a “serious trust crisis” could follow if these data practices became public. How Cursor responded Cursor made two concrete changes after the investigation. First, it restricted access to internal tools so that logins require company email addresses. Second, Cursor decided to stop using employee-generated data to train models. Notably, there has been no publicly confirmed evidence that c...