House probes Airbnb and Anysphere

Airbnb and Cursor maker Anysphere just got pulled into a much bigger Washington fight over Chinese AI. On April 29, the House Homeland Security Committee and the House Select Committee on China opened a joint investigation into both companies, asking how they used Chinese-developed models, what data may have touched those systems, and who inside the companies made those calls. The immediate issue is software choice. The real issue is whether core American products are quietly building on models that lawmakers think come with censorship, security, and supply-chain risks. (homeland.house.gov) Why these two companies? Because each one had a very concrete tie to a Chinese model. The committees’ letter to Anysphere zeroes in on Cursor’s Composer 2, which lawmakers say was reportedly built on Kimi from Beijing-based Moonshot AI. Airbnb got flagged because its customer-service agent used Alibaba’s Qwen model — the one Brian Chesky had praised as “fast and cheap.” This was not a random fishing expedition. Congress picked two visible examples of U.S. companies shipping real products on top of Chinese AI. (homeland.house.gov) What are lawmakers actually worried about? Two things are getting bundled together. First, the classic national-security fear — if American companies send prompts, metadata, or product context into Chinese-linked models, sensitive information could leak or become accessible in ways customers never intended. Second, the newer AI-specific fear — that some of these cheaper Chinese models may have been improved through large-scale “distillation” of U.S. frontier models, meaning American firms would be buying back capabilities that may have been copied from American labs in the first place. (homeland.house.gov) What does “distillation” mean here? Basically, one model learns from another model’s outputs. That can be normal. But the committees and the White House are talking about the hostile version — using proxy accounts, jailbreaks, and evasive access methods to siphon capabilities from top U.S. systems, then repackage them into cheaper products. The April 23 White House OSTP memo said foreign entities, mainly in China, were running “deliberate, industrial-scale campaigns” to do exactly that. So when Congress looks at Moonshot, DeepSeek, MiniMax, or Alibaba-linked systems, it is not just asking where the servers sit. It is asking where the intelligence inside the model really came from. (homeland.house.gov) Why does that matter for Airbnb or Anysphere users? Because the risk is not only espionage in the movie sense. It is dependency. If a booking platform, coding assistant, or enterprise workflow tool starts relying on a third-party model that lawmakers later decide is off-limits, the company may have to rebuild product features, retrain users, and unwind vendor relationships fast. For Cursor, that is especially sensitive because coding tools can touch source code, developer prompts, and internal workflows. For Airbnb, customer-service systems can sit near identity, payments, disputes, and travel records. (semafor.com) Why is this happening now? The timing is not accidental. The probe landed less than a week after the White House memo on adversarial distillation, and it fits a broader congressional push to tighten export controls, punish AI model theft, and keep U.S. companies from normalizing Chinese AI inside mainstream products. John Moolenaar has been pushing that line aggressively all month, and this investigation turns that posture into direct pressure on named companies. (whitehouse.gov) What happens next? The committees asked for information on model selection, communications with model providers, and internal decision-making, and they want relevant employees to attend an in-person briefing. That means this can move beyond headlines into a paper trail — contracts, architecture choices, data-flow diagrams, and procurement logic. Even if no law changes tomorrow, the signal is clear: using a Chinese model is no longer just an engineering decision. In Washington, it is becoming a governance and national-security decision too. (semafor.com) The bottom line is simple. Congress is using Airbnb and Anysphere to test a new boundary for AI adoption in the U.S. If this pressure campaign works, companies across software will have to prove not just that their AI is cheap and good, but that its lineage, safeguards, and data paths are politically survivable. (homeland.house.gov)