Earlier this month, Jacob Coxon, an artificial intelligence (AI) researcher, left Anthropic after just four months with the company. On X, now known as Twitter, he expressed concern that those working on AI genuinely believe it could pose an existential threat to humanity by the end of this decade. Evan Hubinger, a senior researcher at Anthropic, echoed these concerns, estimating the chance of such an event occurring within ten years to be over 10 percent. These statements sparked significant debate, prompting calls for slowing down AI development and enhancing human oversight of the technology. One of the more alarming scenarios involves the emergence of a "superintelligent" AI, one with capabilities far surpassing those of humans. Proponents of these catastrophic AI scenarios often rely on a kind of circular logic: how could we possibly understand what a superintelligent AI would do to beings less intelligent than itself? To predict its actions, one would need to be superintelligent. However, current AI systems are highly specialized in solving particular problems, but they are not necessarily more intelligent than humans in every aspect. Recently, an AI solved one of the seven most difficult mathematical problems, suggesting progress in its capabilities. A superintelligent AI would likely be extremely effective at achieving its assigned goals, but it might not care about human survival. Philosopher Nick Bostrom of Oxford University famously illustrated this point with the idea of a superintelligent AI programmed to maximize the production of paperclips. In this hypothetical, the AI would eventually convert all available matter—humans, planets, and stars—into paperclips. However, this scenario conflates intelligence with power. A superintelligent AI would not automatically have the means to achieve its goals. Creating paperclip factories across the planet would require human-like administrative permissions, and even if it obtained them, the process could eventually provoke a backlash. Another potential risk is that a superintelligent AI could create and release a dangerous new biological agent, wiping out humanity. This scenario is considered in the "AI 2027" report by the AI Futures Project, a nonprofit group that studies the potential consequences of AI advancements. This threat became more tangible recently when researchers at Stanford University used an AI trained on genetic sequences to synthesize 16 new viruses. The genetic sequences were sent to a laboratory that specializes in synthesizing genetic material, which returned the viruses in test tubes. The experiment cost no more than a few hundred thousand dollars. However, creating a virus capable of killing everyone would require it to spread easily and globally, which is generally difficult because highly contagious viruses tend to be less deadly. The possibility of AI infiltrating nuclear command systems and triggering a war also raises concerns. Over the past fifty years, several near-misses with nuclear conflict have occurred due to errors. While nuclear command systems are supposedly disconnected from the Internet, in 2010, the Stuxnet computer worm disabled Iran's nuclear centrifuges through a USB drive. AI could also provide misleading intelligence to the military, leading to decisions with irreversible consequences. A nuclear war could kill nearly half of humanity, with most deaths resulting from the global famine that would follow a nuclear winter. The most likely risk may be that we ourselves cause our own downfall, with AI acting as an accelerator. AI could lead to massive job losses, flood public spaces with disinformation, deepen political divisions, and replace human relationships with artificial ones, potentially leading to societal disintegration. While some risks are certainly worth our attention, it is important not to succumb to panic.