Appearance
The algorithmic state: How the race for military and bureaucratic speed is dismantling human control
WASHINGTON — In an underground command bunker near Tel Aviv, an intelligence officer stares at an amber-tinted monitor. The screen displays a satellite coordinate, a residential address, a probability score of 0.87, and a portrait pulled from a driver's license registry.
The software recommends an airstrike.
Under operational protocol, the human officer must verify the target. In practice, the officer scrolls past the surveillance summary, checks for nearby school locations, and hits approve.
The entire process takes 22 seconds.
"The algorithm does not pull the trigger," a former Israeli intelligence officer said, speaking on the condition of anonymity because they were discussing classified operations. "It simply places the target on your desk with the authority of mathematical certainty. You would need extraordinary courage to look at a statistical model trained on millions of data points and say, 'I know better.'"
Across the globe, the relationship between the state and the citizen is undergoing its most radical transformation since the dawn of the nuclear era. Driven by geopolitical paranoia, venture capital billions, and the seduction of bureaucratic efficiency, governments and armed forces have quietly outsourced the machinery of life, death, and civic survival to algorithmic decision systems.
The shift is not merely technological. It is an institutional abdication, reshaping human psychology, public morality, and political power at machine speed.
Table of Contents
The laboratory of the Donbas
Nowhere has the velocity of autonomous warfare accelerated faster than across the cratered fields of eastern Ukraine.
When the war began in 2022, consumer drones piloted by troops wearing video goggles were celebrated as tactical equalizers. Today, that model of warfare is largely obsolete.
The 600-mile frontline is blanketed by Russian and Ukrainian electronic warfare networks that flood radio frequencies with static. The jamming breaks the radio link between pilots and their drones within seconds of launch.
To survive, the machines had to be taught to think for themselves.
Ukrainian defense startups, backed by the government’s Brave1 innovation cluster, now deploy drones equipped with miniature edge-computing microprocessors running lightweight computer-vision models.
A pilot launches the drone toward an approximate GPS coordinate. Once the aircraft approaches the combat zone, human control drops off entirely. The onboard neural network scans the terrain below, recognizes the thermal signature and geometry of a Russian T-72 tank or an artillery piece, locks onto the object, and steers the warhead into the target at 90 miles per hour.
"When radio frequency is jammed at 200 meters, a human pilot is useless," said Mykhailo Fedorov, Ukraine’s minister for digital transformation. "Without autonomous optical guidance, the drone crashes into the dirt. With it, the machine finishes the kill."
The automation is no longer confined to the sky.
In early 2026, Ukraine deployed more than 12,000 unmanned ground vehicles across the front line. Small, treaded robots armed with remote machine guns now hold defensive trenches in the Donbas, firing at incoming infantry based on automated acoustic sensors and optical tracking.
To refine these algorithms, Ukraine in March 2026 began sharing its combat archive—comprising millions of hours of raw, annotated battlefield video—with defense contractors in the United States and Europe. The bloodiest trench warfare of the 21st century has become the premier training dataset for Western commercial AI.
Russia has matched the pace.
Russian forces now deploy mass-produced Lancet and Kub loitering munitions equipped with autonomous optical seekers. In August, an independent United Nations monitoring mission documented a strike in the Kharkiv region where a Russian drone’s automated targeting system locked onto a white civilian passenger van, misclassifying the roof rack as a troop transport platform.
Seven civilians died. There was no human operator behind the controls to notice the child safety seat in the back window.
The industrialization of the kill chain
While frontline combat uses AI to solve tactical problems, strategic headquarters are using algorithms to revolutionize the scale of violence itself.
During its campaigns in Gaza, the Israel Defense Forces deployed an interconnected suite of artificial intelligence systems named "The Gospel" (Habsora) and "Lavender."
Previous military campaigns produced targets through painstaking human intelligence: analysts cross-referenced phone records, drone video, and informant testimony over weeks to build a single strike package.
The Gospel turned that artisanal process into an assembly line.
Feeding on continuous mass surveillance—intercepted WhatsApp messages, location pings from cell towers, drone video feeds, and social network graphs—Lavender scored tens of thousands of Palestinian men against a predictive profile of suspected militant membership.
According to investigative reporting by the Israeli publications +972 Magazine and Local Call, the system generated target lists containing up to 37,000 individuals at its peak.
Human analysts assigned to oversee the software described feeling like a biological bottleneck in an automated machine.
"I was putting my signature on dozens of targets every day without having the time to understand where the intelligence came from," another former Israeli intelligence officer said. "The machine made the error rate feel acceptable. If the algorithm had an 85% accuracy rate, you reasoned that taking down 10 militants was worth the two innocent people the model got wrong."
A third automated system, colloquially known by military personnel as "Where’s Daddy?", tracked flagged individuals until they entered their family residences, triggering an automated alert for an airstrike. The result was that many targets were struck not while engaged in combat, but at night, surrounded by spouses and children.
The United States has watched these operational proofs of concept with intense interest.
Under the Pentagon’s Chief Digital and Artificial Intelligence Office, the military has institutionalized the Maven Smart System—the successor to the controversial Project Maven launched in 2017.
Integrated across all combatant commands, Maven fuses satellite feeds, commercial synthetic-aperture radar, and intercepted communications into a single interface. At the 18th Airborne Corps in Fort Liberty, North Carolina, commanders use the software to cut the "kill chain"—the time required to find, fix, and strike an adversary—from hours to minutes.
The strategic imperative behind this push is China.
Under the Pentagon’s Replicator initiative, the military plans to field thousands of expendable, autonomous drone swarms across the Taiwan Strait by late 2026. War games conducted by the Center for Strategic and International Studies have consistently shown that human-directed communications would be overwhelmed by Chinese missile batteries and electronic jamming within the first 48 hours of conflict.
In the calculus of the Pentagon, only a self-organizing, autonomous swarm can survive.
The gold rush: From 'Don't be evil' to 'American Dynamism'
The rapid weaponization of artificial intelligence required a fundamental cultural and financial realignment inside Silicon Valley.
In 2018, the tech industry fancied itself a moral check on military power. More than 4,000 Google employees signed a petition protesting the company’s involvement in Project Maven, prompting executives to cancel the contract and declare that Google would never build AI for weapons.
Eight years later, that anti-military consensus has evaporated.
In its place has emerged a hawkish, lucrative alliance between defense procurement officials and venture capital firms under the banner of "American Dynamism." Venture capital investments into defense-tech startups exceeded $14.6 billion in the first five months of 2026 alone, up from less than $2 billion in 2019.
Leading the charge is a new generation of defense contractors who view software, not steel, as the primary weapon of war.
Anduril Industries, founded by virtual reality entrepreneur Palmer Luckey, has raised billions to build autonomous surveillance towers, underwater drones, and interceptor aircraft powered by its proprietary Lattice operating system.
Palantir Technologies, co-founded by billionaire Peter Thiel and led by chief executive Alex Karp, has seen its market valuation surge past $60 billion as its software platforms became the de facto operating system for military intelligence from Kyiv to Taipei.
"The peace dividend of the 1990s is over," Karp told investors during an earnings call in May. "The world is dividing into democratic nations that understand that technology must defend civilization, and those who naively believe that moral superiority alone stops an adversary’s drones."
Behind the rhetoric lies raw commercial self-interest.
Traditional defense giants like Lockheed Martin and Raytheon spent decades securing lucrative "cost-plus" contracts that rewarded slow, bloated hardware development. Silicon Valley startups, accustomed to venture-style rapid iteration, realized that software-defined defense systems offered profit margins exceeding 40%.
Former Google chief executive Eric Schmidt emerged as the ultimate power broker of this transition.
Through his Special Competitive Studies Project (SCSP)—a well-funded policy institute modeled on the Cold War-era Rockefeller Special Studies Project—Schmidt spent five years lobbying Congress and the White House to strip regulatory barriers and redirect federal procurement toward AI startups.
Schmidt invested personally in Ukrainian drone manufacturers, advised the National Security Commission on Artificial Intelligence, and frequently shuttled between Silicon Valley venture firms and the Pentagon’s E-Ring.
The financial gravitational pull eventually bent even the most high-minded AI labs.
In January 2024, OpenAI quietly scrubbed the blanket prohibition against "military and warfare" from its acceptable use policy. By early 2026, OpenAI formalized enterprise agreements with the Department of Defense to deploy its models across non-classified logistics, administrative, and cybersecurity environments.
The friction, when it occurred, was brief and brutal.
When rival startup Anthropic insisted on keeping contractual terms that prohibited the Pentagon from using its Claude models for autonomous lethal targeting and mass surveillance, the Defense Department pushed back aggressively. In February 2026, the Pentagon briefly classified Anthropic as a "supply-chain risk," signaling to federal contractors that ethical reluctance would be treated as unpatriotic obstruction.
Within weeks, federal agencies redirected tens of millions of dollars in planned contracts toward competing labs willing to sign agreements authorizing use for "all lawful national security missions."
The automated clerk: Domestic policing, borders, and the quiet purge
The tools developed for foreign battlefields inevitably turn inward.
While public attention focuses on autonomous drones and military strikes, artificial intelligence has quietly established an automated administrative regime over millions of civilian lives.
Along the rugged borderlands of Arizona and Texas, U.S. Customs and Border Protection has erected more than 400 autonomous surveillance towers built by Anduril.
Equipped with radar, infrared cameras, and computer-vision algorithms, the solar-powered towers scan the desert 24 hours a day. The software distinguishes between cows, deer, and human beings walking miles away, automatically plotting GPS coordinates and dispatching Border Patrol agents via automated dispatch apps.
At international airports, CBP's Traveler Verification Service matches passengers' live faces against Department of Homeland Security photo galleries with an automated confidence score. For American citizens, opting out requires asserting a legal objection to an armed federal officer; for foreign visitors, it is mandatory.
The domestic deployment of predictive software extends deep into the civilian welfare state, where the consequences are less visible than an airstrike but no less devastating.
In Michigan, state officials sought to cut administrative costs by deploying the Michigan Integrated Data Automated System (MiDAS) to detect fraudulent unemployment claims.
The software operated without meaningful human review, scouring tax filings, employer reports, and application forms for discrepancies. Between 2013 and 2015, MiDAS flagged more than 40,000 citizens for fraud, automatically intercepting their tax refunds, garnishing up to 25% of their paychecks, and levying quadruple penalties.
Subsequent legal reviews revealed that the algorithm had an error rate of 93%.
"People lost their homes, filed for bankruptcy, and had their credit ruined by an automated program that flagged them because their employer spelled their name wrong on a tax document," said David Blanchard, an attorney who represented affected residents in a class-action lawsuit that resulted in a $20 million settlement. "When victims called the agency, the staff told them the computer had ruled, and there was no mechanism to overturn it."
In Allegheny County, Pennsylvania, child welfare authorities implemented the Allegheny Family Screening Tool (AFST), a predictive risk algorithm designed to assist caseworkers in deciding which allegations of child neglect warrant formal investigation.
The model calculates a child's risk score by analyzing administrative data, including Medicaid records, food stamp usage, juvenile justice interactions, and county mental health clinic visits.
Because wealthy families use private healthcare and private therapists, their struggles never appear in county databases. The algorithm, by design, flags the poor.
"The algorithm acts as an institutional laundromat for bias," said Virginia Eubanks, author of Automating Inequality and a professor of political science at the University at Albany. "It takes historical patterns of systemic poverty, runs them through a mathematical formula, and produces a score that social workers treat as scientific truth."
At the Internal Revenue Service, artificial intelligence has been deployed to sift through millions of annual tax filings. While the agency boasts that machine-learning models helped identify $1.3 billion in unpaid taxes from complex hedge funds and digital-asset transactions, civil rights groups note that algorithmic audit models historically audit low-income recipients of the Earned Income Tax Credit at rates five times higher than middle-class taxpayers, simply because automated data matching on wage forms is cheaper than auditing a Cayman Islands shell company.
The psychology of abdication: Why humans love the machine
The rapid adoption of algorithmic authority in war and governance cannot be explained by technological capability alone. It satisfies a profound human desire: the elimination of guilt.
Deciding whether to send a missile into a crowded apartment building or strip an impoverished mother of her disability benefits is an agonizing human responsibility. It carries moral weight, political vulnerability, and personal trauma.
An algorithm offers an escape hatch.
Psychologists refer to the phenomenon as "automation bias"—the tendency of humans to defer to automated systems even when their own senses and judgment contradict the computer’s output.
When a drone operator fires because an algorithm highlighted an object in red, or when an administrative clerk denies Medicaid because an automated screening tool returned an error code, the individual does not feel personally responsible.
"The machine provides moral insulation," said Lucy Suchman, professor emerita of anthropology of science and technology at Lancaster University. "It dissolves the burden of conscience. The soldier says, 'I was following the software.' The programmer says, 'I only wrote the code; the commander chose how to deploy it.' The commander says, 'The algorithm is validated by the Defense Science Board.' Responsibility evaporates into the system."
This psychological detachment was documented extensively among drone operators stationed at Creech Air Force Base in Nevada during the height of the War on Terror. Despite being thousands of miles from the physical battlefield, operators suffered from rates of post-traumatic stress and moral injury comparable to infantry soldiers in combat.
Automated targeting systems promise to solve that institutional problem. By compressing the kill chain and framing targets as statistical outputs rather than living human beings, the software creates a clean, clinical distance between the decision to kill and the act of killing.
Greed reinforces this psychological dynamic.
For government administrators under pressure to cut public expenditures, an algorithm is a magic bullet. It allows an agency to eliminate thousands of civil service jobs while maintaining the appearance of efficient service delivery.
For defense contractors, automated systems unlock recurring subscription revenues: an F-35 fighter jet is sold once, but an AI-driven operating system requires continuous data licensing, security patching, and model updates worth hundreds of millions of dollars each year.
The Geneva stalemate and the arms-race trap
For more than a decade, diplomats have gathered at the Palais des Nations in Geneva under the auspices of the United Nations Convention on Certain Conventional Weapons (CCW) to debate an international treaty banning lethal autonomous weapons systems.
The talks have achieved virtually nothing.
The core dispute centers on the concept of "meaningful human control." A coalition of more than 100 developing nations, supported by the International Committee of the Red Cross, has demanded a legally binding treaty that would prohibit any weapon system that selects and attacks targets without human intervention.
They argue that delegating the decision to take a human life to a mathematical calculation violates international humanitarian law and offends the "dictates of public conscience."
Every major military power—the United States, Russia, China, Israel, and the United Kingdom—has systematically blocked any legally binding agreement.
Their counter-argument is framed as a prisoner’s dilemma: in a world without mutual verification, no country can afford to tie its own hands.
"If the United States unilaterally restricts autonomous weapons, we are simply guaranteeing that the Chinese Communist Party will dominate the skies and waters of the Pacific," said a senior U.S. defense official involved in autonomous policy negotiations. "A treaty that cannot be verified is not arms control. It is unilateral disarmament."
In September 2026, the UN disarmament forum reached a watered-down, non-binding consensus framework. It categorized autonomous weapons into two tiers: weapons whose behavior is unpredictable would be prohibited, while weapons operating under "appropriate human supervision" would be regulated through national self-certification.
In practice, the agreement changed nothing on the battlefield.
Every military contends that its weapons operate under human supervision, even when that supervision consists of an operator having 15 seconds to abort a strike launched by a swarm of autonomous aircraft.
The arms-race dynamic has effectively foreclosed diplomatic restraint.
When one state deploys an algorithm capable of selecting targets in two seconds, its adversary cannot afford a decision-making cycle that takes five minutes. The speed of the machine dictates the tempo of the entire conflict, dragging both sides toward total automation.
What holds ahead: Flash wars and the black-box state
The converging trajectories of military deployment and domestic administrative automation point toward three fundamental transformations over the next decade.
1. The 'Flash War' and algorithm-on-algorithm escalation
The most acute danger facing international security is the outbreak of an unintentional conflict driven by algorithmic interactions.
In finance, the automated interactions of high-frequency trading algorithms caused the "Flash Crash" of May 2010, wiping out nearly $1 trillion in market value within 36 minutes before human traders intervened.
Military systems are rapidly approaching a similar threshold of speed.
Imagine an encounter in the South China Sea in 2028: an autonomous Chinese surveillance drone approaches a U.S. naval strike group. The U.S. ship’s automated electronic warfare system detects an active radar lock and initiates an autonomous jamming routine. The Chinese drone's onboard neural network interprets the jamming as an offensive act and deploys defensive decoys that trigger an automated point-defense interceptor from a nearby American destroyer.
Within four seconds, two sovereign powers have fired live ordnance at one another without a single human president, admiral, or general being aware that an incident had occurred.
Because neural networks are fundamentally black boxes—systems whose millions of weightings produce outputs that cannot be fully predicted or reversed by human engineers—the interaction of competing military algorithms in an unscripted physical environment creates unpredictable chaos.
2. The emergence of the 'Digital Client State'
The computational infrastructure required to train, deploy, and maintain frontier AI models is beyond the reach of all but a few nations.
Developing a competitive defense AI system requires billions of dollars in advanced semiconductor fabrication, vast electrical grids, thousands of high-bandwidth data links, and access to immense pools of sovereign data.
As a result, national sovereignty in the late 2020s will be defined by computational dependency.
Countries that cannot build their own foundational models will become algorithmic vassal states, compelled to purchase their national security operating systems from either American defense firms like Palantir and Anduril or Chinese state-backed conglomerates like CETC.
These digital alliances will be far more binding than traditional defense treaties.
If a smaller country depends on an American cloud provider to operate its air defense grid, Washington will hold a digital kill-switch over that nation's ability to wage war or defend its borders.
3. The dissolution of the social contract
Inside domestic borders, the widespread adoption of agentic AI systems within civil bureaucracy threatens to alter the nature of democratic governance.
For centuries, constitutional democracies have operated on the principle that state coercion requires justification. If a government audits your taxes, denies your disability pension, revokes your bail, or arrests you at an international transit point, it must provide a human reason rooted in written statutory law.
The citizen retains the right to contest that reasoning before a human judge.
The algorithmic state dissolves this legal compact.
When decisions are made by multi-variable predictive models, there is no written chain of administrative reasoning. There is only a statistical probability.
The citizen is stripped of agency, reduced to an assortment of telemetry—location history, payment records, metadata associations, and demographic markers—processed by software that the government itself does not fully understand and cannot legally disclose due to private contractor trade-secret protections.
"We are sleepwalking into a world where power is exercised without authorship," said Eubanks. "When no human being can be held legally or morally accountable for an act of violence or an act of deprivation, the rule of law ceases to exist. We are left with an autocracy of the code."
The shift toward autonomous power was never mandated by voters, nor was it authorized by an act of international law.
It was built brick by brick, contract by contract, by military commanders desperate for operational tempo, venture capitalists hungry for sovereign returns, and civil servants seeking refuge from the messy complexities of human judgment.
The technology works. The drones find their targets; the audit models locate the errors; the border towers detect the migrants.
The machine is faster, cheaper, and infinitely more disciplined than the frail human beings who created it.
Yet as the final layers of human friction are stripped from the systems of state power, societies are arriving at a threshold where the code can no longer be recalled. When the machine finally takes full command of the state, it will not be because it staged a rebellion. It will be because humans eagerly handed it the keys.