# I'm disappointed Source: https://www.youtube.com/watch?v=N80TzPCHbNg Recap page: https://rapidrecap.app/video/N80TzPCHbNg Generated: 2026-07-30T04:04:07.15+00:00 --- ## The Gist Matthew Berman is deeply disappointed by Anthropic's hypocritical anti-open-source lobbying and misleading safety claims. Ultimately, Anthropic CEO Dario Amodei publicly insists they do not want to ban open weights models, yet constantly frames open source as a catastrophic national security and cyber risk. ## Quick Overview Matthew Berman breaks down the intense ideological battle over open source AI models versus closed source companies like Anthropic and OpenAI. Despite public denials from Anthropic CEO Dario Amodei, the company constantly lobbies against open weights by claiming they present uncontrollable safety hazards and cybersecurity threats. Berman demonstrates why these safety arguments fail, how open source actually enhances security through transparency, and why economic incentives and data distillation make open source a vital counterweight to tech monopolies. **Key Points:** - Jensen Huang and Nvidia published an open letter signed by dozens of tech companies supporting open source artificial intelligence models. - Anthropic was the sole major AI company that refused to sign Nvidia's pro-open-source letter, while simultaneously running media campaigns against open source safety. - The artificial analysis index shows open source models like Kimi K3 ranking right behind top proprietary closed models like Claude and ChatGPT. - Nvidia recently launched the Open Secure Alliance alongside companies like Microsoft, Palantir, and Cloudflare to build cybersecurity tools for open source agents. - Dario Amodei published an essay on July 27, 2026, explicitly stating that Anthropic has never advocated for a ban on open weights models. - Treasury Secretary Scott Bessent posted on July 22, 2026, warning that foreign actors conducting industrial-scale distillation attacks on American IP will face sanctions. - Anthropic recently settled a massive billion-dollar-plus fine because their models scraped copyrighted books without permission. ![Screenshot at 45:05: Dario Amodei's published essay addressing open weights models, where he claims Anthropic never advocated for a ban despite constant lobbying against them.](https://ss.rapidrecap.app/screens/N80TzPCHbNg/00-45-05.jpg) **Context:** The artificial intelligence industry is deeply split between closed source labs like OpenAI and Anthropic, which treat models as proprietary commercial products, and open source ecosystems championed by companies like Nvidia, Meta, and various international labs. This ideological schism has erupted into high-stakes political lobbying, national security debates, and public clashes over safety regulations. ## Detailed Analysis Matthew Berman unpacks the severe double standards and misleading rhetoric employed by Anthropic regarding open source artificial intelligence. While major tech leaders like Nvidia CEO Jensen Huang rally the industry around open architectures and launch the Open Secure Alliance to protect open source agents, Anthropic remains conspicuously silent and actively lobbies lawmakers against open weights. Berman methodically dissects Dario Amodei's July 27 essay, proving that Anthropic's repeated warnings about biological weapons and cyber attacks are designed to establish regulatory capture and protect their high-margin business models. Furthermore, Berman explains Jevons Paradox in AI compute, detailing how falling inference costs will dramatically increase aggregate demand and ensure open source innovation outpaces centralized proprietary control. ### The Pro-Open Source Letter and Anthropic's Silence Nvidia CEO Jensen Huang published a viral letter advocating for open source AI development, signed by nearly every major tech company except one. - Major enterprises including Microsoft, Meta, Google, and SpaceX co-signed Nvidia's letter stating that open models strengthen safety and cybersecurity. - Anthropic stood alone as the only major AI company that refused to co-sign the open source endorsement letter. - While competitors embraced open access, Anthropic actively pitched news outlets on the existential dangers of unconstrained AI models. ![Screenshot at 01:08: The massive grid of tech company logos that co-signed the letter, highlighting Anthropic's conspicuous absence.](https://ss.rapidrecap.app/screens/N80TzPCHbNg/00-01-08.jpg) ### Defining Open Source AI and the Economic Stack Open source software fundamentally reshapes industries by making core foundational technologies publicly available for modification and execution. - Open source means software can be freely used, modified, inspected, and shared, much like Android dominates the global mobile market against closed iOS. - The AI stack is split into distinct layers consisting of chip manufacturers, datacenters, model providers, software infrastructure, and user applications. - Closed source business models rely heavily on selling high-margin proprietary intelligence tokens, creating a massive financial incentive to crush open source alternatives. ![Screenshot at 13:28: The complete AI stack breakdown from chips and datacenters up to applications.](https://ss.rapidrecap.app/screens/N80TzPCHbNg/00-13-28.jpg) ### Debunking the Safety Arguments Against Open Source Anthropic argues that open weights models are inherently unsafe because bad actors can strip away guardrails. - Open source software is historically more secure than closed software because public scrutiny allows thousands of developers to patch vulnerabilities quickly. - Closed proprietary models also suffer from safety bypasses and jailbreaks, proving that security through obscurity is an illusion. - True AI safety is strengthened when more independent researchers have access to models to build defensive cybersecurity tooling. ![Screenshot at 24:40: The comparative whiteboard breakdown of safety arguments for and against open source AI.](https://ss.rapidrecap.app/screens/N80TzPCHbNg/00-24-40.jpg) ### Distillation, China, and Regulatory Capture U.S. government officials and lab executives frequently raise alarms about Chinese AI labs distilling American frontier models. - Model distillation is a standard, legal training process where a smaller student model learns from a larger teacher model to gain efficiency. - Treasury Secretary Scott Bessent warned that industrial-scale distillation attacks crossing into IP theft will face severe economic sanctions. - Anthropic recently paid a billion-dollar-plus settlement for scraping copyrighted books without authorization, exposing their own IP hypocrisy. ![Screenshot at 37:05: The whiteboard illustration explaining the legal and technical mechanics of AI model distillation.](https://ss.rapidrecap.app/screens/N80TzPCHbNg/00-37-05.jpg) ### Analyzing Dario Amodei's Defensive Response Anthropic CEO Dario Amodei published an official response attempting to walk back accusations of lobbying against open weights. - Amodei explicitly claims Anthropic has never advocated for a ban on open weights models. - Despite denying a direct ban advocacy, Amodei constantly portrays open weights as an unacceptable national security risk. - Matthew Berman exposes this contradiction, arguing that portraying open source as dangerous achieves the exact same protectionist outcome as an outright ban.