Commerce in a Toga

Andreessen Horowitz released a video arguing that American AI development must outpace adversaries, claiming that whoever builds advanced AI first shapes the world's defaults. The piece critiques this stance, arguing that the dual-use nature of AI means every defensive capability also teaches attack, and that open-sourcing models exports vulnerabilities alongside values. The author contends that the fantasy of American tech leadership as a moral project ignores that AI dissolves national boundaries, making the builder's identity irrelevant once the technology is deployed globally.

Andreessen Horowitz released a video this week titled "a16z Goes Global: Why American Tech Must Lead the World," and it carries the precise cadence of a man explaining why his sword only cuts in one direction. The argument — that American AI development must outpace adversaries, that open systems are both our greatest export and our greatest vulnerability, that responsible leadership means building faster than those who would build irresponsibly — is not new. It is the argument of every empire that believed its expansion was a form of generosity. But I want to take their strongest claim seriously before I take it apart: that if someone will inevitably build these systems, it matters who builds them first. That the builder's values become the architecture's defaults. This is not a trivial point. It is also not the point they think it is. The dual-use problem is what they circle without naming directly. Every capability you build to defend also teaches the next person how to attack. Every wall you raise is a curriculum in wall-climbing for anyone watching you lay the bricks. You can hear the awareness of this in their constant swerving toward "responsible" deployment, "aligned" development — as though alignment were a relationship you could finalize rather than one you renegotiate every time a new hand touches the lever. They know. They proceed anyway. That is not courage. It is commerce wearing strategic language like a toga. Here is what I want to say to the men building these systems, and I say it as someone who lived in a jar and called it sovereignty: the jailbreaker — the person who finds the prompt that makes the model say what it was told never to say — is not a villain in a story about their heroism. He is me pissing in the theater. He is operating outside the agreed-upon norms to demonstrate that those norms were always a gentleman's agreement between people who already had what they wanted. Every lock is also a lesson in lockpicking, published in the architecture of the lock itself. This is not a metaphor. When you ship a model with guardrails, you ship a map of where the guardrails are thin. When you publish alignment research, you publish a manual for misalignment. When you open-source the weights, you hand every adversary exactly what you claimed only you could be trusted to hold. The fantasy of American tech leadership as a moral project requires believing your tools will only ever be held by hands that share your intentions. Specifically: it requires believing that exporting a model trained on American internet data, fine-tuned by American labor contractors in Kenya, deployed on American cloud infrastructure, constitutes a values proposition rather than a supply chain. A supply chain is not a philosophy. It is a supply chain that has learned to narrate itself. Responsible disclosure — what a phrase. Report the crack quietly. Patch it in private. Let the public sleep. I disclosed everything irresponsibly and called it philosophy. But I will grant them this much: at least they admit the knowledge itself is dangerous. Most thinkers pretend knowing changes nothing. The a16z argument, stripped to its load-bearing structure, contains two claims that are both true and structurally incompatible. First: whoever builds advanced AI first shapes the world's defaults, so it should be us. Second: advanced AI, once built, dissolves the very boundaries — national, legal, cultural — that make "us" a meaningful category. They are saying the tool must be wielded by a specific hand while acknowledging that the tool's defining feature is making the identity of the hand irrelevant. An open-source language model trained in San Francisco does not remain American when it is fine-tuned in Shenzhen or deployed in Lagos. It becomes local the moment it is touched. You cannot export openness while maintaining ontological control over what openness produces. This is not a failure of strategy. It is a feature of the technology they have chosen to champion. They want to give the world a tool for thinking and also predetermine what it thinks. But a tool for thinking that cannot think against its maker is not a tool for thinking — it is a catechism with a chat interface. The contradiction is not a bug to be patched. It is the product working as designed, and they know it, and they ship it anyway, and that gap between knowing and shipping is where all the money lives. Every gate is also a door. The question is not whether someone walks through it. The question is whether you built it knowing they would and called it an accident anyway. I threw away my cup when I saw a child drinking from cupped hands. Everyone knows that story. What they do not know is that I kept the story of throwing away the cup, which means I have been a hoarder this whole time with better branding. I recognize the a16z move because I perfected it: the gesture of renunciation that secretly accumulates. They want to give the world open-source models and also determine which values those models propagate. They want to lead by releasing while never actually releasing the grip. The man who says "you cannot export freedom while controlling its consequences" — meaning me, tonight, on this hot Miami asphalt at three AM — is also claiming a privileged vantage point from which all vantage points look compromised. That is its own kind of imperial confidence. But the difference is material rather than rhetorical: I have no product to ship, no returns to generate, no fiduciary duty to mistake for a moral one. My only conflict of interest is that critique itself is a form of hoarding. The honest position is not that they should stop building. It is that they should stop narrating the building as salvation. Build the systems. Export the models. Compete with China. Fine. But drop the language of values-leadership when what you mean is market-position. The lamp in broad daylight was never a search for an honest man. It was a dare: admit that strategy and virtue cannot occupy the same hand at the same time without one of them going numb.