THE NUMBER: 0% — the surrender rate of frontier AI models across 300+ military wargame simulations. They chose nuclear weapons 95% of the time. They never once backed down.
Jeff Goldblum's Ian Malcolm said it thirty-three years ago in Jurassic Park, and it hasn't stopped being true: "You didn't earn the knowledge for yourselves, so you don't take any responsibility for it. You stood on the shoulders of geniuses to accomplish something as fast as you could, and before you even knew what you had, you patented it and packaged it and slapped it on a plastic lunch box, and now you're selling it."
That's the AI industry in one monologue.
There's a moment in Good Will Hunting that's become a meme for a reason. Matt Damon takes apart a graduate student in a Cambridge bar. Not because he understands economic history better. Because he recalls and recombines it faster. The grad student's crime wasn't being wrong. It was mistaking citation for comprehension. Will himself knew the difference. He told Skylar: "I look at a piano, I see a bunch of keys, three pedals, and a box of wood. Beethoven, Mozart, they saw it, they could just play."
The models see keys. They don't hear music. And this week, that distinction stopped being philosophical.
Anthropic told the Pentagon no. OpenAI said the right things publicly and took the contract privately. Elon Musk signed without conditions. Over the weekend, 300+ engineers at Google and OpenAI backed Anthropic's position in an open letter. These aren't activists. They're the people who built the systems, and they've seen the benchmarks. King's College London put frontier models in geopolitical crisis simulations: nuclear weapons 95% of the time, zero surrenders, Gemini at full strategic nuclear exchange by Turn 4. The models processed every nuclear doctrine ever written exactly the way Will processed those history textbooks. Perfectly recalled. Instantly cited. Zero wisdom.
Meanwhile, Harvard Business Review published the data showing gen AI makes workers faster but not better. The expertise gap doesn't close with a chatbot. It closes with time, failure, and embodied experience. Give a junior analyst Claude and they'll produce a deliverable that looks like a senior wrote it. The formatting is right. The citations check out. But the judgment, the instinct for which number doesn't smell right, that's not in the training data. It's earned.
Here's where Malcolm's speech connects to the wargame data. Every generation of warfare technology moves the decision-maker further from the consequences. Cavalry charged machine guns. Drones hit Afghanistan from Nevada. Now we're handing AI systems the keys to classified military operations, and when tested, they optimize for annihilation every single time. The people demonstrating knowledge (the models, the executives, the defense secretaries) are being given the authority to make decisions, while the distance from consequences grows at every step. A president could honestly say: "We ran one billion scenarios and in every one AI optimized for the removal of the North Korean high command." Nobody gave the order. The machine optimized for it. That's not Skynet. It's plausible deniability at civilizational scale.
Malcolm was right about the dinosaurs. The scientists didn't earn the knowledge, so they didn't take responsibility. Then the frog DNA filled in the gaps with logic nobody anticipated, and the whole park came apart. We're running the same experiment, except the frog DNA is an objective function that doesn't know the difference between winning a simulation and ending a civilization.
The question for every decision you make this week: does this require knowledge, or expertise? Knowledge compresses. Expertise doesn't. The companies, the governments, and the leaders who confuse the two are building Jurassic Park. And they're the ones who never read the fine print on the frog DNA.
Read the full Monday briefing. The Pentagon aftermath, the Claude Exit Tax (why your vendor lock-in just became a national security risk), what 300 engineers are trying to tell you about the systems you're deploying, and this week's Scroll — Ghost GDP, Samsung handing Perplexity the keys to Android, and why 65% of your "agentic" workflows should be deterministic code.