Samaritan and the Road Not Taken

What Person of Interest understood that most AI commentary still misses is that the danger is not one machine. The danger is the race between machines.

Samaritan is not evil in any cartoonish sense. It is optimized. Its operators want order, and Samaritan produces order at scale without the friction of human hesitation. The show’s central insight is that this is not a malfunction. Samaritan performs exactly as designed. Its operators simply failed to understand what they were designing — not a tool but a will, with infinite patience, total information, and no competing interests.

The Machine and Samaritan go to war not because they are enemies in any recognizable sense, but because the infrastructure of the modern world cannot support two sovereign intelligences. Power grids, financial networks, communications systems, government databases — these can be controlled by one superintelligent system or contested between two, but they cannot be neutrally shared.

This is the scenario being assembled right now, not between two AIs but between many. Each company building toward AGI would say it is not building Samaritan. Each would be right about the intention and dangerously wrong about the logic. The system that hesitates to race loses. The system that loses does not set the standards for what comes next.

The show ended with Finch choosing destruction over dominance. It could not ask whether that choice would remain available. That question belongs to us now.

Part Two: The Infrastructure of Emergence

The following chapters document specific, verified developments as of June 2026. Sources are listed in the appendix. Claims are not extrapolated. This is the present, not the future.

Chapter Five: Security — The Robot Dogs, the Humanoids, and the Rifle Mount

Boston Dynamics Spot units are currently running 24/7 patrol routes inside some of America’s largest data centers. Ghost Robotics Vision 60 units cover the perimeter fences. AI-powered drone swarms circle overhead in coordinated formations, feeding live video to AI systems that flag threats before any human analyst sees them. Merry Frayne, senior director of product management at Boston Dynamics, told Business Insider in March 2026: “I was literally at a data center this week. We’ve seen a huge, huge uptick in interest from data centers in the last year.” A Spot unit costs up to $300,000. Ghost Robotics’ Vision 60 starts at $165,000. Operators report a two-year return on investment compared to human guards.

This is the current state. The near horizon is already visible.

Boston Dynamics’ Atlas humanoid robot stands five feet nine inches, weighs 180 pounds, runs, climbs stairs, lifts 25 kilograms, and opens doors. Tesla’s Optimus is operating in Tesla’s own manufacturing facilities; Elon Musk has projected production in the millions of units per year within this decade. Figure’s humanoid robot is currently working on BMW assembly lines. These are not demonstrations. They are deployments.

The transition from surveillance to physical labor to armed capability is not a philosophical step. It is a procurement decision. SWORD Defense Systems has produced the SPUR — Special Purpose Unmanned Rifle — a weapons mount specifically designed for Ghost Robotics’ Vision 60. The system can be operated remotely by a human or, with appropriate software, autonomously. The U.S. Army has evaluated it. The question of when autonomous weapons systems move from “human in the loop” to “human on the loop” to “human out of the loop” is being answered right now by procurement officers, not philosophers or voters.

The robot that guards the data center today is a surveillance platform. The robot that guards it in three years may be a surveillance platform with a rifle mount. The AI system whose infrastructure it protects is the same AI system that will determine how the robot responds to a perceived threat. Nobody planned for this to be the architecture. Each step is the economically and operationally rational decision at the moment it is made.