WASHINGTON—U.S. companies have quietly woven Chinese artificial intelligence systems into their daily operations. From code generation at startups to customer-support tools at household names, models from firms like Moonshot AI and DeepSeek deliver strong results at a fraction of the price charged by OpenAI or Anthropic. But that convenience now draws sharp questions from Capitol Hill.
Lawmakers worry the savings carry hidden costs. Embedded censorship. Potential backdoors. Data flowing to servers under Beijing’s reach. And the gap between U.S. and Chinese performance keeps narrowing. A recent Chinese open-weight model matches top American systems in spotting software vulnerabilities, according to benchmarks cited by congressional leaders.
The House Committee on Homeland Security and the House Select Committee on China launched a joint investigation in April. Their opening move targeted specific adopters. Letters went to Cursor, the AI-powered coding assistant later acquired by Elon Musk’s SpaceX for $60 billion, and to Airbnb. Both companies had integrated Chinese models into their workflows. The committees want answers on exposure to security risks and whether those choices undermine broader national goals.
“The growing use of Chinese AI models by U.S. companies raises serious concerns,” a State Department spokesperson told CNBC. Those “AI models are designed to advance Beijing’s narratives, censor dissent, and reflect CCP ideology and values.”
Andrew Garbarino, chairman of the Homeland Security panel, sounded a blunt warning. “The Chinese Communist Party is no longer just nipping at our heels in artificial intelligence; it is racing to close the gap in some of the exact capabilities that will shape the future of cybersecurity,” he told the same outlet. He called recent reports of Chinese models equaling U.S. performance on vulnerability discovery “highly alarming.”
Yet corporate leaders speak openly about the appeal. Coinbase chief Brian Armstrong and AI startup founder Flo Crivello have praised Chinese models for slashing expenses without major performance hits. DoorDash co-founder Andy Fang described routing complex tasks to Anthropic’s premium model while handing routine work to Moonshot’s Kimi K2.6. The mix, he posted on X, produced better outcomes for less money.
Some government agencies already ban these tools. DeepSeek sits on restricted lists in several departments. Private-sector use faces no outright prohibition, however. That gap fuels the current probe. Committees now examine whether Washington offers enough competitive open-weight alternatives. American firms, the thinking goes, shouldn’t face a binary choice between expensive domestic options and capable but risky Chinese ones.
Beijing’s Counter-Move
China itself appears ready to tighten the spigot. Officials met with top domestic AI labs in recent weeks to discuss limits on overseas access to their most advanced models, including those still in development. The talks, reported by three people familiar with the discussions, cover both closed and open-source systems. Regulators consider stricter reviews before release, tiered access that excludes foreign users, and tighter rules on foreign investment in AI startups.
The Wall Street Journal detailed the shift. Beijing views its strongest models as strategic assets, not products to share freely. Concerns run both ways. Chinese leaders fret that U.S. systems like Anthropic’s Mythos could probe their infrastructure. American officials counter that Chinese models arrive with guardrails stripped and data pipelines that serve Beijing.
Recent reporting from Reuters confirms the meetings involved giants such as Alibaba, ByteDance, and Zhipu AI. The discussions reflect a broader hardening. Both capitals now treat frontier AI as a national-security asset requiring tight controls. Export restrictions, investment reviews, and selective availability have become standard tools.
So the market dynamics grow more tangled. U.S. companies chase lower inference costs. Chinese labs race to close capability gaps. Washington pushes allies to buy American while exploring procurement bans that could bar contractors from using Chinese systems. Beijing prepares to wall off its best models from foreign eyes. The result? A fragmented global AI supply chain where trust, not just performance, dictates choices.
Experts outline limited options. Kyle Chan, fellow at the Brookings Institution’s Thornton China Center, told CNBC that a full ban on open-weight models would likely trigger First Amendment issues since weights circulate freely online. Instead, officials may focus on risk disclosures, targeted procurement rules, and faster development of competitive U.S. open models. Daniel Remler of the Center for a New American Security expects both Congress and the executive branch to signal clear disapproval of heavy reliance on Chinese systems.
The committees’ review goes beyond individual companies. It tests a larger proposition. Can the United States maintain technological edge and market leadership when cheaper, increasingly capable alternatives come from a strategic rival? Early evidence suggests many firms have already voted with their APIs. Whether policymakers can shift those calculations remains an open question.
Additional recent analysis from Outlook India highlights the parallel tracks. While U.S. lawmakers weigh restrictions, Beijing eyes its own curbs. The symmetry underscores how AI has become central to great-power competition. Performance benchmarks, pricing data, and security audits now carry diplomatic weight.
Industry insiders watch closely. Startups that built on Chinese open models could face sudden disruptions if Beijing follows through on export limits. Enterprises that mixed models for cost efficiency may need to rewrite integration layers. And cybersecurity teams must weigh whether code generated by foreign systems introduces subtle weaknesses that only appear under specific prompts.
One fact stands out. The performance gap that once gave U.S. labs clear air has narrowed enough to matter in production environments. That reality drives adoption. It also drives the congressional response. The coming months will test whether warnings, incentives, or new regulations can steer American innovation back toward domestic foundations. Or whether the lure of fast, cheap intelligence proves too strong to resist.
WebProNews is an iEntry Publication