The Fight Over Open Source AI, Anthropic's $1.5B Payout, NYC Socialists: Evictions = Violence?
David Sacks, David Friedberg, Chamath Palihapitiya, Jason Calacanis
24 Jul 20266 min read1h 30m
TL;DR
China's Moonshot AI released Kimi K3, an open-source model matching GPT-5.6 at ~50% lower cost, triggering White House discussions about banning Chinese open-source models. David Sacks, Chimath, and Friedberg argue this would be a catastrophic own-goal: American companies would pay a 'token tax' while the rest of the world uses cheap open-source AI freely, and Anthropic — which grew from $10B to $70B ARR this year — is accused of regulatory capture rather than genuine national security concern. The real fix, they argue, is forcing Anthropic to KYC its customers to stop distillation at the source, not punishing the entire American open-source ecosystem.
Key Moments
David Sacks
“The tell on this, the way that you know that this whole distillation thing is fake is because if stopping distillation was their primary objective, Anthropic would push to ban Chinese access to American models, not American access to Chinese models.”
Sacks making his sharpest argument that Anthropic's lobbying is about competitive protection, not national security.
“These models are getting commoditized much faster than anybody thought. And how do we know this? Because there is no meaningful sustained advantage once a model publishes their performance criteria.”
Chimath explaining why closed frontier lab valuations are structurally at risk as open-source models match them within weeks of launch.
“At Google in the early days, we would submit millions of search queries to Yahoo and Microsoft search engines to see what the result sets were and we would compare our results against their results as a way of improving our search engine rankings and our algorithm.”
Friedberg drawing a direct historical parallel to normalize AI distillation as standard competitive benchmarking, not theft.
“OpenAI and Anthropic have both argued that they are free to train on all the world's output regardless of whether the creator wants them to or not. That is their current position.”
Sacks pointing out the hypocrisy of Anthropic and OpenAI complaining about distillation while themselves training on third-party content without permission.
All-In is a weekly podcast hosted by Jason Calacanis, David Sacks, David Friedberg, and Chamath Palihapitiya — four tech investors and operators who debate the biggest stories in tech, business, and politics. Known for its candid takes and insider perspectives, the show has become one of the most influential voices in Silicon Valley. The hosts frequently clash on policy, markets, and strategy, making for genuinely unpredictable conversations.
Takeaways
1
Distillation ban logic is backwards If Anthropic genuinely wanted to stop Chinese distillation of its models, it would lobby to ban Chinese access to American models — not ban Americans from using Chinese open-source models. The fact that it's pushing the latter reveals the real goal is eliminating a cheaper competitor, not protecting national security. Sacks calls this the definitive 'tell' that the distillation argument is pretextual.
2
Anthropic could stop distillation with KYC Distillation at industrial scale — where millions of accounts are created to query models and feed outputs into training — is inherently detectable. Implementing KYC (know your customer) with real ID verification would substantially reduce it. The hosts argue Anthropic hasn't done this because it would slow account growth and revenue, meaning the company is prioritizing ARR over the national security threat it publicly warns about.
3
US open-source ban = forced AI token tax If the US government restricts American enterprises to Anthropic and OpenAI only, those companies could charge 50-100x what open-source alternatives cost. That cost differential shows up in corporate earnings, forces Wall Street to rerate US companies downward versus global competitors using cheap open-source AI, and ultimately tanks the broader stock market. Chimath calls it 'regulatory capture' masquerading as national security policy.
4
Sacks: White House has not decided to ban open source Despite Axios reporting the White House was 'considering' a ban and Polymarket pricing it at 45% probability, Sacks says he has it on 'good authority' that no decision has been made and the administration wants that known. Howard Lutnick reportedly opposes a ban and instead favors incentivizing US labs to produce better open-source models. The palace intrigue suggests the administration is still in an advisory-gathering phase, making outside voices relevant.
5
Distillation is industry-standard benchmarking, not theft Friedberg notes that Google routinely submitted millions of queries to Yahoo and Microsoft in its early days to benchmark and improve its own search rankings — the same logical structure as AI distillation. The distinction that matters legally and ethically is whether you are copying model weights (the software itself) versus observing outputs and learning from them. OpenAI and Anthropic have themselves argued they can train on any publicly available output, making their objection to Chinese distillation logically inconsistent.
6
Frontier model value is shifting to infra and apps Kimi K3 matching top American models — even partially — within weeks of launch illustrates that foundational model performance is becoming a commodity at extraordinary speed. Chimath argues capital markets will eventually price this in: the sustainable margins are in cloud infrastructure (as Google's earnings confirm) and application-layer businesses, not in selling model tokens. Frontier lab valuations priced at 25-50x open-source alternatives are increasingly hard to justify.
7
Open-source internet precedent predicts AI outcome Netscape built a proprietary browser and server and was eventually crushed when Mozilla, Firefox, Apache, and Chrome open-sourced the web stack. That shift didn't destroy the internet economy — it created Google, Amazon, eBay, and millions of small businesses by making infrastructure free. Friedberg argues open-source AI follows the same trajectory: value diffuses to applications and cloud, not to the small group controlling the model gate.