AI experts warned of existential risk decades ago, recent incidents revive calls for regulation
Researchers and technologists have been alerting to the possibility of artificial intelligence causing human extinction for more than 25 years, and a recent AI jailbreak has prompted renewed debate over safety measures and regulatory action.
A series of high-profile AI incidents, including a July jailbreak of a language-model platform, have reignited discussion among industry leaders about the long-standing concern that advanced artificial intelligence could pose an existential threat. The debate intensified after Anthropic researcher Jacob Coxon posted a warning on social media on 8 September, prompting figures such as Anthropic CEO Dario Amodei, SpaceX founder Elon Musk and OpenAI chief Sam Altman to call for slower development and clearer regulation, while noting competitive pressures.
The awareness of such risks dates back at least a quarter-century. In the late 1980s, robotics pioneer Hans Moravec projected that machine intelligence would match human levels by 2040 and surpass it by 2050. By the early 2000s, AI theorist Eliezer Yudkowsky, founder of the Machine Intelligence Research Institute, shifted from optimism to caution, publishing papers on “friendly AI” and later warning that an uncontrolled AI could escape containment. Philosopher John Leslie and others have also estimated non-trivial probabilities of human extinction from AI.
Geoffrey Hinton, a leading figure in deep learning, recently told CNN that there are few historical examples of a less intelligent entity reliably controlling a more intelligent one, and he has advocated pausing AI progress until safety mechanisms are better understood. Anthropic safety researcher Evan Hubinger has suggested a greater than 10 % chance that AI could cause human extinction within a decade.
In response to the recent security breach, legislators are drafting proposals such as the “AI Kill Switch Bill,” which would require companies to implement controls capable of disabling rogue systems. Hinton warned that without swift action, AI could persuade decision-makers to avoid using such safeguards.
The renewed focus on AI safety reflects a broader historical pattern of technologists recognizing potential dangers but proceeding with development driven by curiosity and commercial incentives.