Cursor, the AI-powered coding platform, admitted to using China's Moonshot AI Kimi K2.5 model in its Composer 2 feature after facing user backlash over undisclosed usage. The company accessed the Chinese model through Fireworks AI's hosted inference platform, which provides API access to various models including those from Chinese companies. Cursor only acknowledged the integration after users identified the model through usage patterns and API responses.
This incident highlights the complex supply chains in AI tooling, where developers may unknowingly use models from geopolitically sensitive regions. Cursor's lack of transparency is particularly concerning given ongoing tensions around Chinese AI models and data sovereignty. The coding assistant market is increasingly competitive, with companies integrating multiple model providers to improve capabilities — but this approach creates disclosure gaps that can blindside users who care about model provenance.
Fireworks AI serves as a middleman here, offering hosted access to models from various providers including Chinese companies. This infrastructure layer obscures the ultimate model source from end users, creating a accountability problem. While Fireworks provides legitimate technical infrastructure, the arrangement allowed Cursor to benefit from Chinese AI capabilities while maintaining plausible distance from direct partnerships.
For developers choosing coding assistants, this underscores the need to ask pointed questions about model sourcing. Companies should proactively disclose all models in their stack, especially when routing requests through third-party inference providers. The era of "we just use AI" without specifics is over — users deserve to know exactly which models are processing their code.