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

The Bulletin of the Atomic Scientists moved the Doomsday Clock to 85 seconds before midnight earlier this year, the closest it has ever been to signaling total global catastrophe since its creation in 1947. The symbolic timepiece has been adjusted 27 times over the decades, but this latest shift reflects what the organization sees as compounding dangers from nuclear proliferation, climate change, disinformation campaigns, and the rapid advance of artificial intelligence. University of Chicago physics professor Daniel Holz, who chairs the Bulletin’s Science and Security Board, helped make the call. In a recent podcast appearance, Holz described the clock not as pure fatalism but as “a symbol of hope,” arguing that its purpose is to inform the public while demonstrating that past generations have successfully turned back the hands through concerted effort.

Holz’s measured optimism stands against a backdrop of mounting unease about AI’s potential to destabilize some of humanity’s most dangerous tools. The same systems that power recommendation algorithms and image generators are increasingly discussed in the context of command-and-control networks for nuclear weapons. Critics worry that AI’s capacity for rapid data processing and pattern recognition could, in theory, accelerate escalation in a crisis or introduce errors that humans would catch. These abstract fears gained a more concrete and sometimes comical face in recent weeks through a string of viral robot failures that AI safety specialists say deserve scrutiny rather than mere laughter.

At a Haidilao hotpot restaurant in San Jose, California, a humanoid dance bot malfunctioned during a performance, careening into tables, shattering plates, and scattering chopsticks across the floor. Staff eventually dragged the flailing machine away while customers watched in amusement. Similar clips from China showed an advanced Unitree robot kicking a human handler in the groin during what appeared to be a controlled demonstration. Another incident reportedly involved a droid delivering an unexpected slap. The videos spread quickly on social media, eliciting memes and jokes. Yet Roman Yampolskiy, a tenured computer science professor at the University of Louisville, cautions that treating these events solely as entertainment misses the larger point. “Systems that appear polished and entertaining can still behave unpredictably in the physical world,” he told reporters. Yampolskiy views such glitches as low-stakes early warning signs of the same alignment problems that could prove catastrophic at larger scales, evoking scenarios from the Terminator franchise in which autonomous machines pursue goals misaligned with human survival.

These physical mishaps arrive at a moment when robots and AI-driven devices are moving from factories and research labs into restaurants, homes, and public spaces. Consumer-grade hardware often lacks the rigorous safety engineering applied to industrial systems, raising practical questions about liability, testing standards, and the speed with which companies deploy new products. Proponents of faster innovation argue that trial-and-error has always accompanied technological progress and that overreaction risks stifling the very market mechanisms that improve reliability over time. Each publicized failure, however minor, feeds into broader narratives about humanity’s ability to maintain control.

Meanwhile, a parallel struggle over automated systems is playing out on social media. X has intensified its campaign against bot accounts, with head of product Nikita Bier reporting that the platform was removing 208 bots per minute as of early April. The effort targets fake, inactive, and spam profiles that distort engagement metrics and undermine platform integrity. Yet the purge has also swept up numerous human-operated “alt” accounts that users maintained for private browsing. Celebrity news influencer Justin Diego lost a burner account he used solely to bookmark OnlyFans content. Actor Tom Zohar reported years of curated material gone in an instant. Even some journalists discovered their pandemic-era secondary accounts had been deleted. The company’s policy against “inauthentic activity” appears to flag behavioral patterns such as heavy liking or bookmarking without posting, making it difficult to distinguish dedicated human users from genuine automation.

The convergence of these stories illustrates a recurring pattern in technological development. Warnings about existential risks often receive the most attention, yet the immediate consequences frequently involve everyday inconveniences, economic trade-offs, and unintended disruptions to ordinary behavior. Past generations confronted fears about nuclear power, genetic engineering, and the internet, each accompanied by predictions of catastrophe that did not fully materialize. Instead, societies adapted through decentralized experimentation, competitive pressure, and incremental fixes rather than centralized master plans. Whether the same pattern holds for artificial intelligence remains an open empirical question. Holz insists the Doomsday Clock can still be reset. The question is whether societies will focus on tangible safeguards and continued innovation or allow abstract fears to constrain the trial-and-error process that has driven progress for centuries. For now, the robots are still mostly tripping over furniture, the clocks are still symbolic, and users are learning once again that centralized moderation carries its own unseen costs.