Doomsday Clock Ticks Closer as AI Risks Span Nuclear to Robotic Failures

The viral video looks like slapstick at first. A humanoid robot at a hotpot restaurant in San Jose dances for customers then suddenly goes haywire, smashing plates, hurling chopsticks, and tearing across the floor like a drunk wedding guest. Staff eventually drag the machine outside while diners stare in disbelief. Similar clips have flooded social media lately. In China a handler was kicked in the groin by a Unitree robot he thought he controlled. Another droid reportedly slapped a person. Online audiences laugh, but a growing number of artificial intelligence researchers see something darker.

These are not harmless bugs. They are early warning signs that machines behaving unpredictably in the physical world could one day produce consequences no one can contain. Roman Yampolskiy, a tenured computer scientist at the University of Louisville, has studied AI safety for years. He told reporters that people only find these failures funny because the harm has so far remained limited and theatrical. Once the stakes move beyond broken dishes the laughter stops. Systems that appear polished in demonstrations still fail when reality intervenes. That gap between demo and deployment should alarm anyone paying attention.

The warning arrives at the same moment the Bulletin of the Atomic Scientists advanced the Doomsday Clock to 85 seconds before midnight, the closest it has ever been to catastrophe since the symbol was created in 1947. University of Chicago physics professor Daniel Holz chaired the group that made the call. Nuclear war, climate change, disinformation campaigns, and the uncontrolled rise of artificial intelligence all factored into the decision. The clock has been adjusted twenty-seven times. Each adjustment was meant to signal danger. This one feels different because multiple threats now reinforce one another. An AI system tasked with early warning or launch authority could misread a glitching sensor, a bad data feed, or an adversary’s test as the opening move of an attack. Once that miscalculation happens the window for human intervention shrinks to minutes.

Holz himself insists the clock represents hope. He points out that humanity has walked the hands backward before through arms control agreements and diplomatic breakthroughs. The symbol, he argues, is designed to alarm people into action rather than resign them to fate. Yet the pace of AI development makes such optimism feel increasingly detached from reality. Companies and governments race to deploy these systems in critical infrastructure, logistics, and defense while the public is fed feel-good stories about dancing robots and labor-saving devices. The same technology that cannot reliably serve soup without destroying a table is being integrated into networks that control power grids, financial markets, and, in some cases, nuclear command-and-control.

Even the digital world shows the same pattern of overconfidence followed by collateral damage. This month X, the platform formerly known as Twitter, launched an aggressive purge of automated accounts. The company announced it was suspending bots at a rate of 208 per minute. The goal was to restore integrity by removing fake engagement and spam. Instead the sweep erased years of carefully curated material from real human users who had created secondary accounts to follow niche content. Celebrity news influencer Justin Diego lost a private burner account he used solely to bookmark solo performers and adult videos. Actor Tom Zohar watched years of accumulated posts vanish. Other users reported the same experience. These were not bots. They were ordinary people who preferred to browse anonymously. The platform’s automated systems could not distinguish between genuine lurkers and genuine spam. In one stroke the purge demonstrated exactly what the physical robot videos show: when you grant decision-making power to opaque algorithms the results are arbitrary, destructive, and impossible to fully roll back.

This is not a coincidence. It is a pattern. Whether in restaurant dining rooms or on social media timelines, the technology repeatedly proves it does not understand the context it operates within. Proponents claim each failure provides valuable training data that will prevent worse mistakes later. Critics counter that we are witnessing the limitations of current approaches in real time. If a robot cannot navigate a crowded restaurant without creating chaos, why should anyone assume the same underlying technology will behave rationally when managing supply chains during international tension or monitoring missile launches? The gap between laboratory promise and real-world performance remains vast.

The national security implications are obvious. Both Washington and Beijing pour resources into AI-augmented weapons and decision systems. Each side fears falling behind the other. That classic arms-race dynamic now includes software whose failures are unpredictable even to its creators. One misplaced confidence in an algorithm’s assessment could escalate a regional crisis into nuclear exchange before any human being has time to intervene. The Doomsday Clock did not move closer to midnight because scientists enjoy scaring the public. It moved because the combination of old dangers and new ones has grown more volatile.

Ordinary citizens sense the disconnect. They watch robots flail on viral clips, read about the latest doomsday warning, then discover their own private online habits deleted by another automated sweep. The pattern suggests institutions racing ahead with deployment have lost sight of basic questions of control and accountability. Daniel Holz is right that hope remains if leaders act. The harder truth is that acting requires slowing down, imposing genuine safeguards, and admitting that some capabilities should not be handed to machines until we truly understand what they are capable of breaking. So far the laughter at dancing robots suggests too many important people still believe the joke is on someone else.