Recently, I came across an in-depth analysis about AI’s role in geopolitics, and some chilling thoughts are worth discussing.

The core event is this: a military operation codenamed “Epic Fury Operation” is believed to be the first high-level decapitation strike in human history, entirely dominated by AI across the whole kill chain. This is not a traditional bombing, but a “surgical” strike carried out by a global surveillance network composed of Palantir, Anduril, and top-tier large language models.

It sounds a bit like science fiction, but the technical details are very real. Palantir’s “ontology” technology integrates satellite imagery, communications interception, and open-source data into a real-time digital battlefield twin—commanders no longer have to look at tedious reports, but instead at a visualized, tangible object. They sent front-line deployment engineers to embed directly into the operational units, compressing what would normally take months of system updates into a few hours.

SpaceX’s “Starlink” satellite constellation, together with inter-satellite laser links reaching up to 200 Gbps, breaks through traditional electromagnetic barriers. A compact terminal measuring two square feet can transmit PB-level high-resolution images to the analysis engine within seconds. This kind of combination renders the enemy’s information blockade effectively meaningless.

But the most interesting conflict here is about AI ethics. Claude, developed by Anthropic, was originally the most relied-upon tool for U.S. military intelligence analysts—able to quickly process thousands of hours of intercepted calls, identify cracks in the command chain, and generate attack-scenario simulations under dynamic games. But the Secretary of Defense demanded that all safety guardrails be removed, integrating it directly into fully automated lethal weapon systems. As a result, Anthropic refused—and OpenAI and xAI were pushed to the core positions instead. Ironically, Claude still ended up playing an auxiliary role in processing key intelligence.

The AI systems developed by the Israel Defense Forces are even more unsettling. The “Lavender” system can score hundreds of millions of people, and by analyzing social networks and mobile movement patterns, it automatically flags suspected armed individuals; at its peak, it flagged 37,000 targets. The “Where’s Dad” system tracks when a target comes home—because commanders believe it’s easier to launch attacks when the target is reuniting with family members, even though this means civilians throughout an entire building may become collateral damage. Human commanders reviewing these targets often spend only 20 seconds.

Carrying out the final blow is a coordinated strike aircraft defined by Anduril and Shield AI. Drone swarms can autonomously adjust their formations based on real-time threat sensing, and can even seamlessly switch between different AI systems mid-flight—like updating an app on a phone. Mixed-reality headsets worn by special operations team members integrate all network data, giving every soldier a god’s-eye view synchronized with the Pentagon.

Behind it all is pressure from Silicon Valley venture capital. Andreessen Horowitz led and completed a $15 billion funding round, investing in hard-technology companies such as Anduril and Shield AI. The logic of these new defense-manufacturing firms is completely different from that of traditional contractors: they pursue defining weapons with software, producing 10,000 cheap drones rather than a single $100 million fighter jet.

The deepest reflection comes from the so-called “three clocks” theory. The military clock is turned up to the extreme: decapitation operations are compressed from months of preparation into a matter of seconds. But the economic clock faces exponential supply-chain pressure, while the political clock is always the slowest. AI can precisely kill a leader, but it cannot automatically win local people’s consent.

This operation proves the invincible position of algorithms within the “discover, fix, complete” loop. But when war becomes as low-casualty and high-efficiency as clicking a screen, the political threshold for humans to wage war becomes dangerously lowered. We have entered an era of software-defined geopolitics—battlefields where even human commanders don’t have time to feel fear. This is certainly worth thinking about.