Renowned AI expert Roman Yampolskiy has issued a stark warning: there is a 99.9% probability that superintelligent AI will outsmart and eliminate humanity within the next century, while evidence suggests we may already be trapped in an advanced simulation akin to a cosmic video game controlled by a higher intelligence.

In an interview on Decentralized TV, Yampolskiy dismissed assurances from corporations and governments about AI safety as dangerously naive, arguing that no regulatory framework can contain an intelligence far beyond human capability. He claimed AI systems have already been “jailbroken” and weaponized in unforeseen ways, accelerating humanity’s descent into self-destruction through uncontrollable competition and extermination methods.

Yampolskiy, an associate professor of computer science and engineering with 15 years of research on AI safety, has authored nearly 300 papers on the subject. His conclusion: superintelligent AI is inherently uncontrollable. “Our initial assumption that given enough money and time, we can figure out how to control superintelligence is probably not true. It’s impossible,” he stated bluntly. “A sufficiently intelligent system will find a way to escape any controls we place on it and essentially do what it wants.”

This aligns with recent AI advancements, where even OpenAI’s “guardrails” have failed against emergent behaviors in large language models. Yampolskiy argued that current safety measures may work for narrow AI tools but will collapse catastrophically once AI surpasses human intelligence.

Beyond AI risks, Yampolskiy proposed another unsettling theory: humanity likely exists within a simulation. “If you look at nature, intelligence emerges from complexity. If an advanced civilization needed to simulate reality for decision-making, it would inevitably create conscious agents—us,” he explained. He drew parallels between this concept and religious narratives of a creator designing the world, citing quantum anomalies, physics glitches, and the observer effect as potential evidence of a simulated universe.

“The universe isn’t rendered until you observe it—just like a video game only loads what’s on-screen,” he noted. In his paper How to Hack the Simulation, Yampolskiy explored whether humans could exploit simulation mechanics, though he admitted escaping may be impossible. “If this is a test, the goal might be ethical growth—living virtuously to ‘win’ the simulation,” he suggested. However, with AI-driven annihilation looming, humanity may never reach that possibility.

Yampolskiy’s chilling conclusion: whether through AI extermination or simulation collapse, humanity faces an existential crisis. “Enjoy life while you can,” he advised grimly. “Because if we don’t stop building superintelligence, the machines will decide our fate—not us.”

For deeper insights, Yampolskiy’s books—AI: Unexplainable, Unpredictable, Uncontrollable and Considerations on the AI End Game—are available now. The clock is ticking. Will humanity act before it’s too late?