Cursor May Phase Out Chinese AI Models: What It Means for You
A House report says Cursor plans to explore phasing out Chinese open-weight models. Composer 2 was built on Kimi K2.5. What changes for Cursor users and teams.

A House report says Cursor plans to explore phasing out Chinese open-weight models. Composer 2 was built on Kimi K2.5. What changes for Cursor users and teams.

On November 20, 2026, the House Homeland Security Committee and the House Select Committee on China released a joint investigative report ↗ on American companies using open-weight AI models from China. The investigation began in April with letters to Anysphere, Cursor's developer, and Airbnb, and added DoorDash in July.
The part that matters for Cursor users is one sentence in the full report (PDF) ↗: "Anysphere informed the Committees that it plans to explore phasing out PRC-developed open-weight models because of geopolitical risk." The report also says Anysphere built Composer 2, "the flagship proprietary model in its Cursor coding platform, on a Moonshot AI base model, a fact that became public only after an independent developer identified the underlying model." Its table names that base as Kimi K2.5, used since March 2026.
CNBC ↗ got the report ahead of publication. It notes that SpaceX completed its purchase of Cursor in August at an implied $60 billion equity value, and that Cursor, SpaceX and DoorDash did not respond to requests for comment. Note the wording: "explore," with no timeline.
The image above is an original graphic made by aitoolranked.
The report uses three companies as case studies, and Cursor is the only one described as planning to move away:
Airbnb runs customized Alibaba Qwen models for customer support, including an assistant handling about 80,000 interactions a day.
DoorDash said 25 Chinese model families had been "authorized and deployed in a non-trivial manner," and that replacing one embedded model could take up to six months.
Anysphere described vendor risk controls, and then the planned phase-out.
The committees make 12 recommendations, including a national American open-weight AI initiative, a "model provenance standard and an AI bill of materials," and keeping Chinese-developed models out of federal systems and contract work. The Homeland Security Committee release ↗ lists all 12. None of this is law yet.
Replying to a post about the CNBC story, Elon Musk wrote ↗: "We aren't serving open source models." That reply has had more than 6,600 likes.
Cursor's models page ↗, as we read it on November 21, lists its own models as Grok 4.7, Grok 4.6, Grok 4.5 and Composer 2.5. Third-party options listed there come from Anthropic, Google, OpenAI and Meta. The page does not mention Kimi, GLM, DeepSeek or Qwen. The report does not say what Composer 2.5 is built on. So the open question is less "will Cursor drop a Chinese model from a menu" and more what sits under its own models.
"Is the base model under your coding agent now a supply chain risk?" — Sai Teja ↗ (@saiitoshii)
"Cheap AI models come with a cost that doesn't show up on the API bill." — Satya Srivatsav ↗ (@Sathya9911)
"Your coding tool's supply chain just became a political decision." — Naeem Anjum ↗ (@thisnaeem)
"Long-term: a plan under review, not a cutoff or a price catalyst today." — Paradise Wire ↗ (@xParadiseWire)
Reddit was not reachable from our tools for this story.
Nothing changes in your editor today. There is no announced removal date, and the models page does not list Chinese open-weight models as options. If you use Cursor on a Plus or Premium plan, keep working; this is a policy signal, not a product change.
Teams with procurement rules should ask Cursor directly. The report's main point is that a model's origin can be "several layers beneath what a customer sees." If your company, a government contract or a regulated client needs to know the lineage of every model in your toolchain, ask Cursor in writing what Composer 2.5 is built on. If the report's provenance rules ever become requirements, every coding tool built on open weights will get the same question.
How it compares. Claude Code runs on Anthropic's own models by default, so the lineage question is simpler there. But the cheap end of the market leans on Chinese open weights, from Kimi K3.1 Pro to Tencent's Hy4, and the report concedes the US "lacks a competitive open-weight alternative." For the trade-offs, see our GitHub Copilot vs Cursor comparison, cheaper Claude Code alternatives and best open-source LLM guide.
Verdict: worth watching, but not a reason to switch tools. The bigger story is that model lineage is becoming a compliance question, and Cursor is the first major coding tool to be asked about it in public.