The moral high ground is getting crowded, and it’s starting to look like a firing squad.
Recent protests from Google employees and the hand-wringing over Anthropic’s policy shifts aren't the noble stands they pretend to be. They are symptoms of a profound misunderstanding of how power works in the physical world. When engineers sign petitions to limit military AI because of strikes in Iran or ethical "fallout," they aren't saving lives. They are outsourcing the defense of their own freedoms to adversaries who don't have an HR department, let alone a "Responsible AI" committee.
The lazy consensus says that AI is a tool for peace that must be shielded from the "taint" of defense contracts. The reality is that neutrality in a technological arms race is a myth. By refusing to build for the Pentagon, Silicon Valley isn't preventing war. It’s ensuring that when the next conflict happens, the side with the most sophisticated tech will be the one that doesn't care about your ethical frameworks.
The Anthropic Paradox and the Myth of Pure Alignment
Everyone is up in arms because Anthropic—the supposed "safety-first" darling—is softening its stance on military work. Critics call it a betrayal. I call it a collision with reality.
I’ve sat in rooms where millions were burned chasing "perfect alignment," only to realize that an AI that refuses to assist in defense is just an AI that has been pre-coded for obsolescence. The idea that you can build a transformative, general-purpose intelligence and then put it in a glass box where it only solves "nice" problems like crop yields or climate change is a fantasy.
Defense is not a side project; it is the fundamental infrastructure upon which a stable society—and a stable market—rests. If the West's premier AI labs retreat into a shell of academic purity, they leave the field wide open for state-sponsored entities in authoritarian regimes to set the global standard.
Why the "Limit Military AI" Movement is Flawed
- Precision is an Ethical Imperative: The loudest voices often argue that AI makes war "easier" and therefore more frequent. They miss the nuance of $P(L)$, the probability of loss. In traditional kinetic warfare, collateral damage is a mathematical certainty due to the limits of human perception and analog hardware. AI-driven systems, if developed correctly, provide the granularity required to minimize non-combatant casualties. Rejecting AI in defense is, ironically, a vote for more "dumb" bombs and higher civilian death tolls.
- The Vacuum Effect: When Google pulls out of Project Maven, the need for that technology doesn't vanish. It just gets fulfilled by lower-tier contractors with fewer internal checks, less talent, and zero public accountability.
- The Intelligence Edge: Modern warfare is moving from the kinetic to the cognitive. If you aren't winning the OODA loop (Observe, Orient, Decide, Act), you’ve already lost. A military that relies on human-speed decision-making against an AI-speed adversary will be forced to use more drastic, less surgical options to compensate for its slowness.
Stop Asking if AI Should Be Armed
The question "Should AI be used by the military?" is the wrong question. It’s like asking in 1940 if the military should use internal combustion engines. It is an inevitability.
The real question is: Do you want the AI that governs the future of global security to be built by people who value transparency and civil liberties, or by those who view tech as a tool for total state control?
When tech workers revolt against defense contracts, they are effectively saying they trust the bureaucrats in foreign ministries more than they trust their own engineering teams to build safeguards. It is a staggering display of "not in my backyard" morality. They want the protection provided by the state's security umbrella while actively trying to poke holes in the fabric.
The Hidden Cost of Ethical Narcissism
I have seen companies blow millions on internal ethics boards that do nothing but produce 60-page PDFs that no one reads. Meanwhile, the actual technical challenges of making AI reliable in high-stakes environments go ignored because the topic is "too controversial" for the board.
This is ethical narcissism. It’s about the workers feeling good about their Slack status updates rather than actually solving the hard problems of kinetic safety.
Imagine a scenario where a state-level adversary deploys an autonomous swarm against a metropolitan center. If your defense systems are hobbled by "usage limits" defined by a group of developers in Mountain View, you don't get a moral victory. You get a tragedy.
True expertise in AI safety doesn't mean avoiding the military; it means being deeply integrated with it. You cannot build a "kill switch" or a "safety rail" for a system you refuse to touch.
The False Choice Between Anthropic and the Pentagon
The drama surrounding Anthropic’s policy change ignores the technical reality of the current "frontier" models. These models are already dual-use. You cannot separate the logic required to optimize a supply chain from the logic required to optimize a logistics train for an armored division.
By pretending there is a clean line between "commercial" and "military" AI, the industry is lying to itself.
- Logistics: Is an AI that routes trucks for Amazon "clean," but the same AI routing fuel trucks for the Army "dirty"?
- Translation: Is a real-time translation tool for tourists "ethical," but the same tool for an intelligence officer "evil"?
- Computer Vision: Is a drone that spots cracks in bridges "safe," while a drone that spots an IED "violent"?
The tech is the same. The only difference is the user. By trying to ban the user, tech companies are setting themselves up for a losing game of cat-and-mouse that will eventually lead to heavy-handed government intervention. The state will eventually take what it needs. It’s better for the industry to be a willing, critical partner that can negotiate the terms of use rather than a hostile witness that gets its doors kicked in by the Defense Production Act.
The Hard Truth for the "Not in My Name" Crowd
If you work at a tier-one AI lab, you are already part of the military-industrial complex. Your taxes, your infrastructure, and the very internet you use to organize your protests were born from defense spending.
Claiming the moral high ground now is like a baker refusing to sell bread to a soldier while living in a castle the soldier is guarding.
The downside of my approach is clear: it’s messy. It requires engaging with the reality of violence, the ambiguity of geopolitical interests, and the potential for misuse. But the alternative—abandoning the field to those with no qualms—is a guaranteed catastrophe.
Practical Steps for a Post-Protest Era
Instead of signing petitions to cancel contracts, the industry needs to pivot toward Technical Defense Transparency.
- Embedded Safety Teams: Don't just hand over an API. Embed safety engineers directly with the defense teams to ensure the model’s thresholds for "certainty" are tuned for high-stakes environments.
- Red-Teaming for Conflict: Use the best minds in alignment to simulate how these systems fail in combat scenarios. This is where real "responsible AI" happens—not in a cafeteria debate.
- The "Dual-Use" Dividend: Use the massive R&D budgets of the defense sector to solve the core problems of hallucination and logic that the commercial sector is too cheap to fix on its own.
The era of the "neutral" tech giant is over. You are either building the tools of the future for your side, or you are making it easier for the other side to win.
Stop pretending your code doesn't have a side. Pick one. Be the adult in the room who understands that a sophisticated defense is the only reason you have the luxury of an "ethics" department in the first place.
Build the shield, or get out of the way of those who will.
