In the Book of Genesis, the serpent promises knowledge without consequence.
"You will be like God," it whispers. The fruit is beautiful, the logic seductive: why depend on slow, fallible human judgment when something faster, more precise, more intelligent can decide for you?
We are standing before that tree again. Only this time, the fruit is an algorithm, and the garden is a battlefield.
The War That Changed the Question
On February 28, 2026, the skies over Iran lit up with nearly 900 strikes in just twelve hours. The US and Israel launched a massive preemptive offensive, Operation Rising Lion, using the most advanced AI-integrated warfare the world had ever witnessed. Autonomous drone swarms, satellite-guided by SpaceX's Starlink and Starshield networks, operated with what military analysts called "a large level of independence." AI systems assessed intelligence, identified targets, and simulated battle scenarios in real time.
For the first time in history, the question was no longer theoretical: What happens when machines make decisions about who lives and who dies?
But this question had already been answered quietly, horrifically in Gaza.
Lavender, Where's Daddy, and the Arithmetic of Acceptable Death
Before the 2026 Iran war made autonomous warfare a headline, Israel had already conducted what may be the largest experiment in AI-directed killing in human history. According to investigations published by +972 Magazine and confirmed by The Guardian, the Israeli military deployed an AI system called Lavender that assigned every person in Gaza a score from 1 to 100, rating their likelihood of being a militant. Out of 2.3 million people, at least 370,000 were placed on a "kill list."
The system achieved what its creators considered an acceptable accuracy rate: 90 percent. In cold mathematics, that meant approximately 37,000 people were mistakenly marked for death, and this margin of error was deemed operationally acceptable.
But Lavender did not work alone. It operated in concert with "Where's Daddy?", a tracking system that monitored individuals' mobile phone locations and notified operators when a marked target returned home. The combined workflow was devastatingly efficient: Lavender marked individuals. Where's Daddy tracked them to their family homes. And the bombing commenced typically at night, when entire families were asleep.
The "human in the loop" was reduced to what one source described as a rubber stamp, spending roughly 20 seconds confirming each target before authorizing a strike. Twenty seconds between an algorithm's recommendation and a family's annihilation.
This is not science fiction. This already happened.
Anthropic vs. The Pentagon: A Battle for the Soul of AI
Against this backdrop, a legal battle unfolding in San Francisco courts carries far more weight than corporate litigation. It may be the defining moral confrontation of the AI age.
In July 2025, the Department of Defense awarded contracts worth up to $200 million each to Anthropic, Google, OpenAI, and xAI. The Pentagon wanted unrestricted access to AI models across all military applications. Anthropic, the maker of Claude, agreed to adapt its policies for defense use, but drew two clear red lines: no mass domestic surveillance, and no fully autonomous weapon systems.
The response was unprecedented. On February 27, 2026, President Trump directed federal agencies to "immediately cease all use of Anthropic's technology." Secretary of Defense Pete Hegseth designated Anthropic a supply-chain risk to national security, a label historically reserved for foreign adversaries like Huawei or Kaspersky. Defense contractors and partners were barred from working with the company.
Anthropic became the first American company to be treated as a national security threat for refusing to build weapons that kill without human oversight.
A federal judge in San Francisco, Judge Rita Lin, granted a preliminary injunction in Anthropic's favor, issuing a 43-page ruling finding the government's actions likely constituted First Amendment retaliation. Yet an appeals court in Washington, D.C., subsequently denied Anthropic's request to block the Pentagon's blacklisting.
The message from the government was clear: if you build powerful AI and refuse to let it be weaponized without constraint, you become the enemy.
I believe, personally and deeply, that this dispute is not merely legal or commercial. It is a battle between good and evil for the future of artificial intelligence. It is the most consequential ethical confrontation since the Manhattan Project scientists debated whether to demonstrate the atomic bomb or drop it on a city.
The Vanishing Human
The Pentagon's Replicator initiative, expanded with $1 billion in funding, aims to deploy thousands of small, expendable autonomous drones. Palantir's Maven Smart System, now operational with NATO, is expected to begin transmitting "100 percent machine-generated" intelligence to combatant commanders by mid-2026.
Consider what this means. When AI systems fuse satellite imagery and drone feeds to surface targeting recommendations faster than human analysts can process them, the nominal "human in the loop" becomes what researchers call an approval node, present in name but incapable of meaningfully interrogating a process they did not design and cannot fully audit.
The edge-cloud architecture of modern autonomous weapons means the human may participate in cloud-based mission planning but be entirely absent from the terminal engagement, the moment when the weapon meets its target. The decision to kill becomes distributed across a network of algorithms, sensors, and pre-programmed parameters. No single human pulled the trigger. No single human can be held accountable.
This is the architectural elimination of moral responsibility.
Asimov Warned Us. We Turned His Warnings Into Entertainment.
Isaac Asimov understood this problem seventy years before it materialized. His Three Laws of Robotics-"a robot may not injure a human being, must obey human orders, and must preserve its own existence"-were never presented as a solution. They were presented as a thought experiment about failure. Every Asimov story is fundamentally about the ways these seemingly airtight laws break down in practice, creating moral paradoxes that no amount of programming can resolve.
We consumed his stories as entertainment. Hollywood turned them into spectacles: I, Robot, The Terminator, The Matrix. We watched civilizations fall to their own creations with popcorn in hand, never imagining we would be the generation tasked with building the machines.
Ray Kurzweil predicts that the technological singularity, the moment when artificial intelligence surpasses human intelligence in every domain, will arrive around 2045, with artificial general intelligence achievable by 2029. As we approach these milestones, the question is no longer whether machines can make autonomous decisions. It is whether we should allow them to make decisions about life and death.
And critically: once we cross that threshold, can we ever come back?
The Original Sin
This is where my metaphor becomes more than literary. The story of the forbidden fruit is, at its core, a story about irreversibility. Adam and Eve were not punished merely for disobedience. They were expelled from paradise because they acquired something that could not be un-acquired. The knowledge of good and evil, once tasted, could not be untasted. And the consequences fell not on them alone, but on every generation that followed.
The decision to remove humans from the kill chain is the fruit.
Once autonomous weapons are normalized, once military doctrines are built around them, once economies are structured to produce them, once the speed of warfare makes human oversight physically impossible, there is no returning to the garden. Once built, the infrastructure of autonomous killing becomes self-perpetuating. The arguments for keeping it become stronger with each deployment: it is faster, cheaper, more precise, and more politically palatable than sending young soldiers to die.
And when not if these systems fail catastrophically, when they mistake a wedding for a weapons cache or a school for a command center, who answers? The algorithm has no conscience. The operator had twenty seconds. The commander followed the machine's recommendation. The programmer wrote code years ago for a different context. The politician signed a contract.
Accountability dissolves into the architecture itself.
The Singularity Paradox
Here lies the deepest paradox of our approach to the singularity. We are building systems of unprecedented intelligence while simultaneously dismantling the frameworks of human judgment that might govern them. We are racing toward a future where machines can think, decide, and act autonomously, and our first major application of that autonomy is killing.
If this is how we use artificial intelligence in its adolescence, what does its maturity look like?
The 2026 Iran war demonstrated that autonomous systems now operate at speeds and scales that make meaningful human oversight physically impossible in many combat scenarios. Lavender and Where's Daddy demonstrated that even when humans are nominally in the loop, the sheer volume and velocity of AI-generated decisions reduce human judgment to a formality.
As UN treaty negotiations on autonomous weapons stall, specifically because the US and Russia refuse to sign binding prohibitions, we are witnessing the international community's failure to set boundaries before the technology makes boundaries irrelevant.
What Developers and Tech Leaders Must Understand
This is not a distant policy debate. It is the defining question of our industry.
Every developer building AI systems, every CTO making architectural decisions, every startup founder choosing which contracts to pursue, is participating in this choice. The line between civilian AI and military AI is not a wall; it is a membrane, and it is growing thinner every day.
Anthropic's stand and the brutal retaliation it provoked should serve as both inspiration and warning. Inspiration because it proves that principled positions are possible even under extreme pressure. Warning because it shows the cost of those principles: being labeled a national security threat for insisting that machines should not kill autonomously.
The question for every person in technology is not whether this matters. It is which side of this line you stand on when the moment comes.
A Letter to the Next Generation
To those who will inherit the world we are building: I am sorry that we are making these decisions without your consent. I am sorry that the seduction of speed, efficiency, and strategic advantage is so powerful that even well-intentioned people struggle to resist it. I am sorry that by the time you read this, the fruit may already have been eaten.
But I want you to know that not everyone reached for it willingly. That there were companies that said no. Scientists who objected. Writers who warned. People who understood that some thresholds, once crossed, redefine everything that comes after.
The singularity may be inevitable. But the choice of what we bring to it is our values, our restraint, our insistence on human dignity that is still ours to make.
For now.
