You can’t ban open source AI

Open source AI cannot be banned, argues a new analysis, as the technology's decentralized nature makes prohibition impractical. The piece highlights that attempts to restrict open source AI would stifle innovation and drive development underground, citing the global community's reliance on open models for research and progress.

You can’t ban open source AI Monday. Today, we’re taking a longer swing at one topic, so the rest of the missive is compressed up top. To work — Alex P.S. I was on This Week in Tech yesterday https://www.youtube.com/watch?v=cwx6CPkMoLg , if that’s your jam. 📈 Trending Up State violence https://www.nytimes.com/2026/07/19/us/politics/fbi-ice-agents-investigations-shootings.html?smid=nytcore-ios-share … Moonshot’s IPO prospects https://www.theinformation.com/briefings/moonshot-ai-seeks-investor-approval-begin-ipo-process … subscription pricing https://9to5mac.com/2026/07/17/apple-raises-prices-for-apple-music-and-apple-one-subscriptions/ … Nvidia’s competition https://www.cnbc.com/2026/07/20/amd-helios-microsoft-ai-nvidia.html … AI bottlenecks https://thenewstack.io/ai-agent-infrastructure-bottleneck/ … booms, busts https://www.ft.com/content/97eeb736-f8af-4839-8511-3d0354c8b34c … free speech https://www.status.news/p/new-york-times-subpoenas-trump-qatar-air-force-one … AI usage in the UK https://www.ft.com/content/1ab00e18-995a-4cda-b76b-00ba6c043863 … regulatory capture https://marginalrevolution.com/marginalrevolution/2026/07/trial-lawyers-lobby-against-autonomous-vehicles.html … passive partners https://www.wsj.com/finance/venture-capital-anthropic-spark-yasmin-razavi-93dd6041?st=MwcFqN&reflink=desktopwebshare permalink … Kimi K3 inference https://openrouter.ai/moonshotai/kimi-k3 activity 📉 https://finance.yahoo.com/news/servicenow-pledges-1-5bn-investment-110000403.html Trending Down 📉 https://finance.yahoo.com/news/servicenow-pledges-1-5bn-investment-110000403.html … The popularity of state capitalism https://www.cnbc.com/2026/07/20/kalshi-traders-think-coinbases-trading-volume-will-fall-again.html cybersecurity https://techcrunch.com/2026/07/20/hugging-face-confirms-breach-affected-internal-datasets-and-credentials-urges-users-to-take-action/ … oil prices https://www.cnbc.com/2026/07/20/oil-prices-today-brent-wti-crude-us-iran-centcom-hormuz.html ? … Coinbase earnings hopes https://www.cnbc.com/2026/07/20/kalshi-traders-think-coinbases-trading-volume-will-fall-again.html … tariffs https://finance.yahoo.com/markets/article/trumps-temporary-global-tariffs-are-set-to-expire-this-friday-131345766.html … crypto regulation https://www.cointribune.com/en/genius-act-a-year-later-final-stablecoin-rules-still-pending/ ? … Hope for the American talent magnet: News that Moonshot AI’s founder and CEO earned his PhD in the United States only to build his company in China kicked off another round of debate around immigration https://x.com/signulll/status/2077973223929479317 . So when OpenClaw Foundation’s head Dave Morin asked https://x.com/davemorin/status/2078210013592715265 why we “don’t we have the US Department of Recruiting that recruits the best engineers in the world to America,” I chimed in, replying https://x.com/alex/status/2078329783075713026 that “insta nt green cards for any PhD would be insanely powerful.” It was just a throwaway tweet, but I didn’t expect the level of fury my little post sparked. Some folk argue that allowing foreign students to stay in the United States and build should be limited to only STEM graduates, or we’ll risk being “flooded by gender studies PhDs.” That was among the most measured views. Others said allowing foreign students to stick around and start companies would flood the nation with spies ? or collapse American wages. None of the ensuing name-calling aimed at me was very creative, sadly. Woof. A huge number of America’s greatest companies were founded or co-founded by folks born elsewhere: Nvidia, Google, Yahoo, Zoom, Stripe, DoorDash… the list is really, really long. These very companies employ hordes of American citizens, and have built our technology industry into the dominant juggernaut that it is today. But if what I read this weekend on X reflects the national temperature on high-skill/education immigration, we’re in deep shit. Nativism is a disease. Will the United States bork Chinese AI models? Open-weight and open-source AI models are not the same. Some ‘open’ models merely release their weights, the numerical parameters that define how they turn inputs into outputs. Other open models release their source code, training data, and weights. Both are far more transparent than closed-source models, which share nothing. Below, ‘open’ means open-weight or open-source. A brilliant new open model from China. A viral essay gone wrong. Claims of regulatory capture running wild. Government meddling… It’s been an insane few days. If you need to catch up on the madness or simply want to see all that you’ve recently read in a single digest, look no further. Let’s untangle the latest blowup in technology-land, kicked off by the launch of the impressive Kimi K3 https://www.cautiousoptimism.news/china-finally-built-an-american-ai-model/ . 1. Chinese AI lab Moonshot’s new Kimi K3 model is the real deal Rumors that Moonshot “benchmark-overfit” K3 are incorrect, as the “model has raw IQ,” according to private tests by the Vercel folks. Not only does K3 outrank https://artificialanalysis.ai/ all but the latest Anthropic and OpenAI models in general intelligence, it’s also aces at cybersecurity work. Vercel’s CEO Guillermo Rauch writes that “ frontier, open-weight cybersecurity capability is here https://x.com/rauchg/status/2078647648307880209 .” - So much for the lead that Mythos could have offered the West. 2. Open, Chinese AI may pose a threat to American AI labs’ business models Think-tanker and former White House technology policy advisor Dean Ball recently joined OpenAI as its Head of Strategic Futures. He stuck his head in the anthill over the weekend, too. Responding to the release of Kimi K3, Ball made a few points that we’ve condensed below https://x.com/deanwball/status/2078133895766114412 : “Open-weight models are inherently decelerationist as in the end, open-weight models deter further AI capex.” The logic here is simple. It’s very expensive to train new, frontier-level AI models. Closed-source AI models are inference-controlled, in that their developer determines who serves them to customers and at what price. Open-weight AI models can be served by any inference provider. So in a closed-source AI world, value accrues to the model maker. In an open-weight world, value accrues to the compute provider only some, due to the margin destruction that perfect competition https://www.investopedia.com/terms/p/perfectcompetition.asp brings and the end-user who benefits more . Thus, if you can’t generate meaningful revenue from training a new AI model, progress will slow down, demand will soften, and the capex frenzy will fade. So, in Ball’s world, open-weight AI leads to less investment in AI. An “open-weight-model-dominant world” threatens “full AI communism,” where “AI is a ‘public good’ which will ultimately be provided by the state as a kind of ‘digital public infrastructure’.” The logic here goes: If there’s less compute available in the free market, the government may step in to provide it. Shared, state-sponsored compute. Think bread lines for AI inference, if you will. These arguments were wildly unpopular on X. Hermes’ Teknium asked https://x.com/Teknium/status/2078574209845068217 “did you just argue monopolies are accelerationist or what ? ” Cloudflare’s Dillon Mulroy called parts of Ball’s argument “ insane https://x.com/dillon mulroy/status/2078519940106051830 .” And Epic Games’ Tim Sweeney tried to compare the arguments to tacos https://x.com/TimSweeneyEpic/status/2078294024239169639 . 3. The government may wield regulation to disincentivize the use of Chinese models Compared to the unpopularity of Ball’s views on open-weight AI models, his riff on the future of domestic AI policy regarding Chinese AI models was radioactive. Current and former Trump administration staffers sounded off, at times confusing Ball’s statements about what may happen under an open AI world, and what he thinks should happen. Regardless, David Sacks shouted https://x.com/DavidSacks/status/2078826291638522127 that we “are at a critical inflection point in AI policy where leading closed labs, already a duopoly in terms of AI model revenue, want the government to eliminate their open source competition.” There’s more. Former Uber embarrassment https://www.nbcnews.com/tech/tech-news/uber-ceo-s-right-hand-man-emil-michael-leaving-company-n771086 and current Under Secretary of War Emil Michael called Ball a “ village idiot https://x.com/USWREMichael/status/2078924719538262107 ” for noting that several American governmental arms have “blocked their employees from using Chinese AI, citing ill-justified claims of danger.” Per Michael, yes, some models were blocked, but the process was “democratic” and not some “Deep State scheme like Ball would prefer.” In contrast, the blocking of Mythos and the forced slow rollout of GPT-5.6 were done with full transparency. /s In short, OpenAI and Anthropic’s many critics loosely, folks aligned with Elon Musk https://x.com/chamath/status/2078835348830912801 consider Ball’s argument as evidence that American AI labs want to close the door on what could be their most viable competition: cheaper and open models from China. 4. The Trump admin is considering policy maneuvers to limit Chinese AI models Axios reports https://www.axios.com/2026/07/20/ai-us-china-open-source-kimi a series of ideas from the second Trump administration to bolster domestic AI, including: Subscribe to Cautious Optimism to unlock the rest. Become a paying subscriber of Cautious Optimism to get access to this post and other subscriber-only content. Subscribe https://www.cautiousoptimism.news/subscribe/?utm source=Cautious+Optimism&utm medium=website&utm campaign=article-paywall