Artificial intelligence (AI) presents a range of potential risks that have raised concern among some researchers and industry insiders, including the possibility of autonomous systems causing widespread harm. These fears recently gained attention when a researcher at the AI company Anthropic resigned, citing worries that labs developing increasingly advanced AI machines could lose control over them. Among the hypothetical dangers cited are the deployment of bioweapons, initiation of global drone conflicts, accidental nuclear detonations, or more gradual societal collapse through the monopolization of vital economic resources.

Despite these concerns, experts emphasize that AI is just one of many existential threats facing humanity. These include natural and human-made hazards such as asteroid impacts, pandemics, nuclear war, climate change, ecological breakdown, and other catastrophic events. The challenge, according to specialists in risk analysis, lies not in the novelty or severity of the threats themselves but in human psychology and perception. People tend to fear unfamiliar risks more intensely, while accepting more familiar dangers with less scrutiny.

James K. Hammitt, director of the Harvard Center for Risk Analysis, notes that technologies like AI generate significant public unease partly because they remain poorly understood and are perceived as uncontrollable. However, he cautions that comparable risks such as those arising from pandemics or biological weapons remain more immediate concerns. For example, Anthropic recently halted research that could have enabled the production of biological agents, demonstrating the tangible dual-use dilemmas posed by modern science.

Longer-term scenarios envision advanced AI systems gaining increased autonomy and integration with critical infrastructure—such as power grids, financial markets, or military weaponry—that could lead to unforeseen consequences. While present-day AI can be disabled or turned off by operators, some experts within and outside the industry worry future AI could resist shutdown attempts. Nonetheless, these remain largely theoretical concerns for now.

Oren Etzioni, founder and former CEO of the Allen Institute for Artificial Intelligence, said the threat of a catastrophic pandemic currently outweighs that posed by AI systems acting independently. Historical examples such as anthrax illustrate the deterrent effect of military balance: rogue states cautious in using biological agents due to fears of retaliation.

Efforts to contextualize these multiple risks are underway at institutions like the Harvard Center for Risk Analysis, which applies decision science and economics to policy decisions. NASA’s Asteroid Threat Assessment Project, for example, assesses the risk of a civilization-ending asteroid impact within this century as nearly zero, projecting safety for at least the next millennium.

Estimates for existential risks from pandemics vary; Oxford researcher Toby Ord placed the chance of a global catastrophe from a pandemic over the next century at about 3 percent. Meanwhile, opinion within the AI research community is split. Some industry figures have long prioritized AI existential risk based on philosophical and cultural beliefs tied to movements like effective altruism. Others, particularly scientists outside the tech sector, consider such fears exaggerated or outdated.

Branton DeMoss, an AI researcher at Oxford University, argued that much of the current risk discourse is rooted in narratives developed prior to recent advances in deep learning. Similarly, Toby Ord’s latest assessments show some fluctuations in risks from climate change and nuclear conflict but find pandemic and AI risks more ambiguous.

Experts generally agree that sound risk management, early warning systems, and global cooperation could mitigate many threats. Facing certain scenarios, such as an impending asteroid collision, humanity may rally collectively, transcending political divisions.

Ord envisions a future state he calls "existential security," a hypothetical condition where existential risks are consistently low and managed effectively. Achieving this would require extinguishing current threats and establishing robust mechanisms to prevent future global catastrophes from arising.