Forgot your password?
typodupeerror
AI China United States

Lawmakers Probe Growing Use of Chinese AI Models In US Companies (cnbc.com) 109

U.S. lawmakers are probing the growing use of Chinese AI models by American companies, citing concerns over censorship, security risks, and whether U.S. firms are turning to cheaper foreign models because domestic alternatives are too costly or restricted. The investigation is specifically looking at companies such as Cursor and Airbnb. "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." CNBC reports: The House Committee on Homeland Security and the House Select Committee on China said in April they will jointly investigate the growing adoption of Chinese-developed AI models. An initial step in the probe was for the chairmen of those committees to send letters to Cursor and Airbnb, over their "use of or exposure to these risks" through AI developed in China. "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," Andrew Garbarino, chairman of the U.S. House Committee on Homeland Security, told CNBC. "Recent reporting that a Chinese open-weight model can match leading U.S. models in certain vulnerability discovery and cybersecurity tasks is highly alarming," said Garbarino.

While some government departments have banned the usage of Chinese AI models including DeepSeek, adoption of them by U.S. companies is not prohibited. Tech chiefs, including crypto company Coinbase's Brian Armstrong and AI startup Lindy's Flo Crivello, have been publicly touting the use of models from China to reduce costs. Cursor, which will be acquired by Elon Musk's SpaceX for $60 billion, built its Composer 2 model using Chinese AI model Kimi, which was developed by Moonshot AI. Alongside focusing on the rise of Chinese AI models, the ongoing joint House Committees' investigation is also looking into whether the U.S. is doing enough to tackle their rise. "The Committees are also examining whether the United States has a sufficient open-weight AI strategy to ensure American companies and cyber defenders are not forced to choose between expensive or restricted U.S. models and cheap, capable PRC-developed alternatives," a Committee aide, who asked not to be named as they were not authorized to discuss the ongoing probe, told CNBC.

[...] The administration could consider the use of federal procurement bans, which would include restricting government agencies and private companies that serve the U.S. government from using Chinese AI models, Kyle Chan, fellow in the John L. Thornton China Center at think tank Brookings, told CNBC. "However, it's ultimately impossible to ban China's open-source AI models because their model weights are available freely on the internet," Chan added. "This could enter into first amendment speech issues." [...] Another [approach] could be disseminating findings about risks and vulnerabilities associated with Chinese AI models to U.S. companies. "Regardless, I do expect both the Executive Branch and Congress to communicate their interest not to see U.S. companies adopting these models," [said Daniel Remler, senior fellow, technology and national security program at think tank the Center for a New American Security (CNAS), told CNBC].

This discussion has been archived. No new comments can be posted.

Lawmakers Probe Growing Use of Chinese AI Models In US Companies

Comments Filter:
  • by fahrbot-bot ( 874524 ) on Thursday July 09, 2026 @07:42PM (#66230820)

    Lawmakers Probe Growing Use of Chinese AI Models In US Companies

    You use a Chinese AI Language Model and an hour later you want to use it again. :-)

  • by bartoku ( 922448 ) on Thursday July 09, 2026 @07:48PM (#66230824)

    It really seems like OpenAi, Google, Meta, and SpaceXAi would do well to release more open weight models.
    Their current models are still a few months ahead of the Chinese models, so their last versions would be right on par.
    It still takes a lot of expensive hardware to run the truly competitive Chinese models, as it would the on release old American models as well.
    As a result you are not really going to lose much if any of your customers to those running your models locally, because if they are capable of doing that they are already looking else where.

    But sadly the American frontier Ai model companies have mostly regressed in releasing open weights.
    Meta has gone closed and is no longer really even talked much about.
    Grok open weight releases are are behind on the promised release schedule sadly; and generally even behind in terms of their latest frontier model.
    Although it will be interesting to see how the new Grok 4.5, the first model it trained with Cursor, is going to score.

    • by jhoegl ( 638955 ) on Thursday July 09, 2026 @08:48PM (#66230868)
      They want it to depend on their data centers, because that is how they make money.

      Businesses cant operate on their shitty model because it costs too much in resources, time, and actual money.

      Comparatively the human brain can produce better results, knows how to qualify facts and compile information and put it into a consumable format for other humans

      But again, how can they grift off of that? Paying others? Thats not how oligarchs want the world to work. Heel only is their goal. So they shoehorn their terrible business into a requirement that people use it, including putting it in OSes, slapping it on browsers, and pumping it into your phone.

      Because your data for free to them, is valuable to them.

      enjoy.
      • "A sovereign system keeps organizations in control of their data, AI, and decision-making rather than depending on external providers"

        https://www.scry.llc/2026/07/1... [scry.llc]

        Look at the Google trends graph. I posted this before about six months ago, the trend is gaining traction.

        • by jhoegl ( 638955 )
          yup, that is how it SHOULD be used, and many places are doing this. The issue is the "engine" as it were that manages all of that, which is what this is about.
      • by Zocalo ( 252965 )
        Wait, what? They're *making* money now? Last I heard they were still playing shell games with pretend money in a financial merry-go-round of pinky swear deals to make it seem like they are somehow not haemorrhaging quite so many hundreds of billions of dollars as they actually are to try and keep the VC funding flowing in.

        The AI endgame, sure. That's totally going to be the kind of bait and switch that Google pulled when they transitioned from a search provider into an ad provider; get everyone hooked
      • Bond: You expect me to use your data centers?

        Villain: No mister Bond, I expect to make use of your data!

        The cloud model they are pushing is about data collection.

    • by DarkOx ( 621550 ) on Friday July 10, 2026 @09:45AM (#66231438) Journal

      They are counting some combination of legitimate risk, FUD, and protectionism to ultimately protect them from the Chinese models.

      The reality is at some point in the not to distant future it will be cheaper to put enough AI accelerator hardware in workstations to give most folks using Claude/Claude code and similar a perfectly acceptable degree of performance. It always goes this way - it is never cheaper put hardware behind the glass when it can go under the desk long term. The only reasons to do it usual boil down to management and wanting to do something more bleeding edge that hasnt filtered to commodity hardware yet.

      Of course for online applications that need to scale, and for complex engineering or very large data volume tasks, sure "Cloud AI" and certainly for anyone who needs to train a model. However the idea these guys are going to get individuals and business to keep paying $200 for tokens to use some desktop AI assistant is unrealistic, and down goes the datacenter volume requirements along with that.

      Again I am not saying there isnt a new industry / space here or that it is all a bubble but the current Anthropic/OpenAI/Grok business model persisting for a whole lot longer does not appear to me anyway to that it fits the patter of the last 25 years of White-Collar-targeted IT systems.

      Let me caveat that I also think the sorts of people making big investments in Data Centers are not stupid and at least see this as a likely outcome as well, presumably they believe they can sell the space/capacity to other users for other applications. If so why not charge the Anthropics of the world with the VC money huge premiums to rush build outs while you can get them? As long the assets are still marketable after that business drops off, it is a win!

    • by Mal-2 ( 675116 )

      DeepMind has released Gemma 4, which doesn't include their largest models, but 31B is a creditable model. I wish I could say the same about the 26B-A4B model, which is just smart enough to be entertaining for a few days or weeks, then everything it does starts to sound the same. Unfortunately, 31B on my hardware (i5-8500, 48 GB DDR4, 12 GB RTX 3060) runs at 0.3 to 1.2 t/s. So while I'm not a big fan of Alphabet's business practices at large, they aren't regressing. Switching to the MIT license basically mea

      • by bartoku ( 922448 )

        Yeah, out of the US frontier model providers Google seems to be giving us the best open weight solutions lately.
        However, Gemma 4 is probably about a year behind the frontier models, and maybe nine months behind the Chinese open weight models.
        It seems Google could do better than Gemma 4 today, given Qwen3.6 is very useful with similar model sizes.

        Of course Google Gemini has been a distant third now in the frontier race as well.
        I want Gemini 2.5 Pro open weights, last years frontier model, or Gemini 3.0 Pro.
        T

    • Regressing on open source is exactly the plan, pushing the industry toward Chinese models and yelling about national security, and needing government intervention to support their business, that is the playbook.

  • by SeaFox ( 739806 ) on Thursday July 09, 2026 @07:52PM (#66230830)

    ...whether U.S. firms are turning to cheaper foreign models because domestic alternatives are too costly or restricted.

    something, something... free market.

  • As an american... (Score:3, Interesting)

    by DiSKiLLeR ( 17651 ) on Thursday July 09, 2026 @08:14PM (#66230852) Homepage Journal

    As an american... the US models are _garbage_. Full or restricitons and censorship. I can't do anything without triggering some kind of cybersecurity flag and blocking me. Fable 5, Opus 4.8 and even Sonnet are useless. Codex, too. Just blocks you and won't work. We need UNRESTRICTED MODELS.

    > AI models are designed to advance Beijing's narratives, censor dissent, and reflect CCP ideology and values.
    Ironically, this isn't true and it's the US models that censor and restrict. The Chinese open source models are GOLD.

    • Re: (Score:2, Troll)

      Ask it what happened in Tiananmen Square in 1989...

      Of course, being open weights, abliterated is a wonderful word to see after a model name.

      • by caseih ( 160668 ) on Thursday July 09, 2026 @10:42PM (#66230962)

        At what point will approved American models affirm that the 2020 election was stolen?

        • by Bahbus ( 1180627 )

          When there is factual data proving it was. Which there isn't any because it wasn't. But Republicans don't know any better because all Republicans are gullible and retarded.

          • by 0123456 ( 636235 )

            It's basically the IQ bell-curve meme again. The retards and the smart people can see it was obviously stolen. The midwits continue to shriek "CITATION NEEDED!" because they're not capable of looking at the evidence themselves and have no common sense.

            • by Bahbus ( 1180627 )

              smart people can see it was obviously stolen

              Like whom? I certainly haven't seen any.

              they're not capable of looking at the evidence

              Sure would help their case if there actually was any evidence.

              • by 0123456 ( 636235 )

                > Like whom?

                Uh, smart people? I know a lot of them and we tend not to hang out with midwits.

                We knew it was stolen by noon the next day, simply by looking at the stats and alleged voting patterns. It was impossible to legitimately get the results they claimed.

                > Sure would help their case if there actually was any evidence.

                Like I said...

                • by Bahbus ( 1180627 )

                  Uh, smart people? I know a lot of them and we tend not to hang out with midwits.

                  Yeah? Name them, then. Or, at the very least, name their qualifications that define them as "smart".

                  It was impossible to legitimately get the results they claimed.

                  No, it isn't.

                  Like I said...

                  You haven't given any. You haven't linked to any. Because you have *nothing*.

                • Uh, smart people? I know a lot of them and we tend not to hang out with midwits.

                  Actual smart people don't need to claim they're smart, nor do they avoid interacting with their fellow man. It's usually the aforementioned fellow man that says one is smart.

                  We knew it was stolen by noon the next day, simply by looking at the stats and alleged voting patterns. It was impossible to legitimately get the results they claimed.

                  Actual smart people don't just know something; they look at the evidence. They also show their evidence when making claims. Neither of which you have done here. What you have here is a belief, and it's done in spite of the evidence. An orange con man's claims are not evidence.

          • Technically no. Because models are not sources of fact, they're just language models. So while it is factual that 2020 was not stolen, that really doesn't matter to the model. Drop a website claiming otherwise into the context and ask it to summarize, and it will.
            • by Bahbus ( 1180627 )

              They can, and do, try to reference sources of fact. Summaries are not affirmations.

              • Some are designed to summarize search results or other sources. Use the word "fact" with caution. My point was simply that scoring models based on what claims they make without external reference is fraught, because that's not "what they do." Trying to extract facts from the model itself is an ill-posed problem.
                • by Bahbus ( 1180627 )

                  Sure. I'm assuming the general AIs that are connected and have access to the internet (ChatGPT, whatever X AI is called, etc). They can and do "affirm" with external references, which is why (to the OP) they won't affirm the election was stolen because there is no factual evidence to affirm.

          • The funny thing is after Trump won and Democrats started claiming that the elections were rigged [youtube.com]. Of course, somehow Trump fully accepted it that time.

            How to recognize a true fact-denying partisan.
            • by Bahbus ( 1180627 )

              They weren't "rigged" but voters were suppressed, which almost like rigging but not exactly.

              • How to recognize the true at heart partisan.
                • by Bahbus ( 1180627 )

                  How to recognize the true at heart partisan.

                  This is nonsense.

                  • Probably not. Would you ever vote Republican?
                    • What kind of a question is that? Voting for a party is dependent on that party's policies. I've seen many people switch parties in my life to suit the actions of the party of the day. Your question is better summed up as "did you vote republican?", given that there's no election cycle and no idea what the upcoming republican party will look like post Trump Cult.

                      Your post makes it very much sound like you're a partisan hack rather than voting for those who best support your interests. The idea that you would

                    • by Bahbus ( 1180627 )

                      Would I? Sure, it is technically possible. If a sane Republican existed with fair policies, a proven track record, and wasn't motivated by self-interest, I would possible vote for them. I just haven't seen one in roughly 15 years and with the current state of the party, don't see one appearing any time soon.

                      Democrats aren't always better, but they do have a better baseline average of being decent human beings. Elected Republicans are statistically more likely to be criminals (specifically sex offenders) and

                    • Well, I think you are more reasonable than that.

                      There is a segment of the population who will never switch parties under any circumstances. They are the partisans. They are easily recognizable because their opinions always somehow match the platform of the party (which is silly, because a party platform is of necessity a compromise between people with different desires).

                      I've seen many people switch parties in my life

                      These people were not party loyalists. Note that not all party loyalists are braindead: some of them get real benefit from being loyalis

                    • But unfortunately this makes Democrats boring. And most Americans now vote for the politician solely based on how interesting they are. If a politician is racist diddles children and doesn't understand the world.. guess what, more interesting.
      • It's an open model. Search for abliterated models on HF. It's rather easy to take these models and retrain them to go outside original constraints.

        Of course if you are using a model as a service, it's rare to find these models. I think OpenRouter has one provider with an uncensored model. But honestly, if you are playing with these models, you likely don't want your ptompts sent over the internet.

    • by broward ( 416376 )

      prove there's nothing malicious in the Chinese models.

      you can't. it's too much data and too many pathways.

      https://www.scry.llc/2026/05/1... [scry.llc]

    • by Himmy32 ( 650060 )

      Also of note with the Open Weight models there are tools that can strip the censorship layers. [github.com] But the government goal is always going to be the "correct type of censorship".

      • by cowdung ( 702933 )

        So my question is how is DeepSeek censored (when you run it locally)? Was data about Tianamen Square ommitted? Was it fine tuned to censor? Did they touch the neurons directly to turn it off? LoRA?

        • So my question is how is DeepSeek censored (when you run it locally)? Was data about Tianamen Square ommitted? Was it fine tuned to censor? Did they touch the neurons directly to turn it off? LoRA?

          DeepSeek the service / website is very much censored using front end filters. The model itself not so much. You can talk to it about Tianamen until you are blue in the face.

      • You call it "the correct type of censorship", most people call it ethical and legal.
    • by AmiMoJo ( 196126 )

      How about Grok? xAI is being sued because it lets you make child abuse material, which sounds pretty unrestricted.

    • There is a thing called 'abliteration' of models, you know... Abliterated Gemma, GPTOSS, Qwen3.6. Just grab a local abliterated model from Ollama and you can ask it to do anything. It's COMPLETELY uncensored. That is what 'abliteration' does. It targets all the 'No I can't do that' pathways and deletes them. It's like evil Data from STTNG.

      Enjoy.

  • to keep the bubble rising. Next we'll have mandatory AI model usage. Have you fulfilled your quota citizen?
  • I wouldn't be surprised if the Chinese strategy to destroy the US Bubble AI economy is by doing what they are best at which is copying American inventions and then releasing them and selling them cheaply even if they are slightly inferior but cheap enough and good enough.

    They did this with American manufacturing, American EV vehicles, all the products sold made of chinesium on Amazon, and now they just have to continue the steal, distil, copy, and then undercut as a strategy to destroy the American AI compa

    • by HiThere ( 15173 )

      IIUC, the Chinese strategy is to come up with a cheaper collection of tools, including models, that are incompatible with the standard US tooling, and convince other countries to use their version instead. The US government seems to be actively pushing to make that strategy work. (E.g. abruptly cutting off access to models with no warning or explanation.)

      • Incompatible with our tooling? All the models are on OpenRouter and can easily be switched between in seconds. The tooling is the same.

        • by HiThere ( 15173 )

          They've got a lot more components to their system...and the tooling isn't all software. As for software, IIUC, they're using a different multi-processing library. Making the weights compatible doesn't make the entire system compatible.

        • They can be called in the same way, but not all of them are as good at working in a particular harness as each other. Anthropics models have been trained to call tools in a certain way, the chinese models may or may not work in a way that is compatible. Tool calling ability (not just whether or not you can do it, but how and when it decides which tools to call and why) differs WILDLY between open models.
    • I guess at this point, though, selling any microchips to consumers is nice even if they're knockoffs since we can't buy RAM or GPUs anymore
    • by jamienk ( 62492 )

      LLMs are a new and developing set of techniques and computer people should be experimenting. The "speed" of development has been turned into "speed" of trying to BLOCK experimentation. We are being instructed to think of "AI" as something that only a few behemoth companies do. Like how "computers" are just MS and Apple, or "phones" are just AT&T, or "Christians" are just in Rome, or "voters" are only landowners, "writers" are only monks... This article is part of a bald and disgusting POWER-GRAB.

    • American companies aren't "overcharging" for their tokens. They are still offering them at an incredible discount, compared to the actual infrastructure and compute cost - a $200/mo OpenAI subscription can let you burn through about $14,000 of token value.

      American companies are still deep in the "get them addicted, then increase the price" phase. If you think the recent token price apocalypse is bad, wait until OpenAI and Anthropic's loans come due shortly...

      China, as usual, appears to be slightly smarter

  • Qwen (Score:4, Interesting)

    by darkain ( 749283 ) on Thursday July 09, 2026 @09:42PM (#66230908) Homepage

    Qwen works. It does what I need. It does it well. I can run it locally on my own hardware.

    FUCK YEAH I'm using a Chinese model because like hell am I paying per-token costs from the subscription model companies!

    • Re:Qwen (Score:4, Interesting)

      by coofercat ( 719737 ) on Friday July 10, 2026 @09:42AM (#66231432) Homepage Journal

      Fun fact... I'm fiddling about "with AI" a bit, and burned through my free credits on Gemini. To keep the party going, I downloaded a new model into my Ollama - qwen3.

      I wasn't quite sure if everything had restarted and whatnot, so I asked the chat tool "what model are you". It said, and I quote:

      > I am Claude, a large language model created by Anthropic. I'm designed to be helpful, harmless, and honest in my interactions. Is there something specific you'd like to know or discuss?

      I'd literally selected 'qwen3-coder:30b' in the model selector thing in Ollama, and yet it told me it's Claude. I didn't even do any clever 'prompt hacking'.

      So yeah, Qwen works - it works really well, and as it's running locally now, it's 'free' too.

  • If you are going to anthropomorphize a tool enough to call it "AI" I suppose that it isn't entirely unreasonable to suspect that it might "reflect CCP ideology and values"; but fretting along that line seems to either be ignoring or deliberately obfuscating the difference between running a model and suckling on someone's black-box API.

    A local bot can hurt you to the degree that you trust it to actually work; but a remote vendor can(and, given the competitive scramble for training data, almost certainly w
  • by jamienk ( 62492 ) on Thursday July 09, 2026 @11:57PM (#66231030)

    US companies like OpenAI and Anthropic WANT regulation in order to block weaker players from developing competing AIs because they can't conform to the regulations. The "costs" of running AI will be the cost of having massive data centers with all kinds of anti-copyright regulations, and a proliferating list of other phony safeguards, including "safe guarding" the companies' own interests in parallel with being moral police, fear mongers, and buck-passers. Chinese and independent developers will train on pirated data and will not need data centers because they will be able to run locally. This will give them some competitive advantages - so the US companies will push to make training data, i.e, all writing and reading, be regulated – no one will be allowed to read websites unless they are logged in and ID authenticated. US AI companies will pay for access. Non-payers will be accused of violating international trade pacts, etc. Their AI models will be banned and there will be a movement to consider them criminal, selfish, etc. The main foreign policy of the US will be to get other countries to agree to license training data, and this issue will dominate US policy. When I (me, a programmer) want to program/train my own AI, it will be made illegal or at least very difficult. Right now it is difficult from a tech resource and org resource POV. But soon all the difficulty will be strictly from a regulatory POV. My access will increasingly need to be monitored and blocked.

    But all of this is just a continuation of the DRM fights and the Free Software battles of the last decades. It isn't even a huge shift. It is a disgusting and short-sighted vision of our values. The worst parts of our culture and business practices have come to dominate the world.

  • Hey Congress (Score:4, Insightful)

    by Snotnose ( 212196 ) on Friday July 10, 2026 @12:02AM (#66231032)
    How's that "let's not let anyone sell advanced chips to China" working out for you? From where I sit it looks like not so well.
  • If access to US models can be switched off by executive caprice (viz Fable 5 and Mythos 5)*, anyone remotely concerned about building a reliable system is almost obliged to use a non-US model. Doesn't hurt that the Chinese ones are good, and cheap, but they win on the commercial consideration no one is going to switch business on and off at random.

    *Some will say "that was only an foreign export restriction," but if that power exists it will be used against anyone inconvenient.

  • All the US companies are captured by Israelies. I wouldn't trust my company to use Anthropic or Pantone foras long as I'm alive
  • Imagine all the money they are saving by not using US AI rates!

    Sorry I meant, this is a clear security breach of very important stuff.
  • by Targon ( 17348 ) on Friday July 10, 2026 @07:45AM (#66231258)

    Here in the USA, there is a fairly large group of people who are against higher education overall, as well as politicians who are worried about religion not being attractive for those with a greater amount of education. They claim that students are being indoctrinated against religion by higher education, but then they push for religion to be pushed in public schools. The fact that there isn't a HUGE push to teach math and science in the USA is a big part of why the USA has lost most of its edge. Those who are stupid but have a lot of money have also pushed a anti-education agenda because they feel intimidated by those who are more intelligent as well as more knowledgeable than they are.

    • by cowdung ( 702933 )

      US high school education prioritizes sports and high school musicals.

      Great for winning Olympics and other sporting events. Great for actors and Youtubers.

      Not great for science. And now the science strongholds in Universities are being threatened by funding cuts, and by blocking the qualified immigrants from other countries that were helping prop it up having had much more demanding high school educations in science and math.

  • While I did expect that this product would be made cheaper and better by China eventually, I did not expect that to happen so fast.

    • What do you think Altman did that Chinese companies couldn't do. Commit mass piracy? No China is the best at that.
      • by gweihir ( 88907 )

        I did expect the hardware limitations and more rational stance would slow them down more. I was mistaken. I see no special accomplishments in Altman besides being an accomplished manipulator and fraudster.

        • Sounds like you fell for the words of another fraudster.
          • by gweihir ( 88907 )

            And who would that be?

            • Trump is doing everything he can to impede on their progress but he doesn't quite get anyone who isn't a wealthy white male.
              • by gweihir ( 88907 )

                And now please explain why I did fall for the words of Trump.

                Because hardware limitations (export controls, most critical for the later ASML machines, and in place longer than Trump) are real. And a "more rational stance" refers to China being much more competent in furthering their agenda and LLM-type AI is clearly not nearly as good as the hype suggests. I do not see Trump in there anywhere...

  • by OzJimbob ( 129746 ) on Friday July 10, 2026 @11:03AM (#66231554) Homepage

    "Companies turn to cheaper, better products with the functionality they want, US lawmakers outraged".

    Seriously, this is getting embarrassing now. Can't buy Chinese routers, can't buy a Chinese EV, now can't use a Chinese LLM? The rest of the world is getting on with things while America will become more of a backwater than it is already, and they will wonder why...

    • Thanks to our great leaders, we're being given freedom from choice. We no longer have to suffer from choice paralysis, just use what you're told.
  • i fully support an end to infinite scroll i think autoplay is ok in some situations
  • As more products and services have software and bots developed in unfriendly nations, the US banning them could put the USA at an economic disadvantage to nations that allow them. I'm not disputing its a legitimate concern, only saying the tradeoff may grow harsher over time.

    We are already paying much more for EVs than we otherwise could. (Although part of that is from trade disputes rather than concerns over rogue XiWare.)

Are you having fun yet?

Working...