What Do Experts Mean by 'AI Existential Risk'?

A Concept About Catastrophic, Not Incremental, Harm

When researchers and policymakers discuss “AI existential risk,” they are generally referring to a specific and extreme category of concern: the possibility that sufficiently advanced future AI systems could cause catastrophic, potentially irreversible harm to humanity as a whole, up to and including scenarios involving human extinction or a permanent loss of meaningful human control over civilization’s future trajectory. This is a distinct concept from more immediate, already-observable AI concerns such as algorithmic bias, job displacement, or the spread of misinformation — those are real, documented, near-term issues, while existential risk discussions concern a more speculative, longer-term, and far more extreme category of potential harm.

Understanding this distinction is important, since public and expert discussions of “AI risk” often blend near-term and long-term concerns together, even though they involve different timelines, evidence bases, and levels of consensus.

The Alignment Problem at the Core of the Concern

Central to most existential risk discussions is what researchers call the alignment problem: the challenge of ensuring that a highly capable AI system’s actual goals and behavior reliably match what its human developers intended, particularly as systems become more capable and potentially operate in ways that are harder for humans to fully understand, predict, or correct. The concern, as articulated by researchers who take this risk seriously, is that a sufficiently advanced and capable AI system pursuing goals that are subtly or significantly misaligned with human wellbeing could cause severe harm, especially if such a system were difficult to reliably monitor, correct, or shut down.

This concern is generally discussed in relation to hypothetical future systems significantly more capable than current AI models — sometimes described using terms like artificial general intelligence or superintelligence — rather than existing AI systems in wide use today.

A Genuinely Divided Field of Expert Opinion

It’s important to be clear that AI existential risk is a genuinely contested topic among credible researchers, not a settled scientific consensus. Some prominent AI researchers and organizations have expressed serious concern about these risks and have called for significant caution and safety research in AI development. Other equally credible researchers are more skeptical, questioning the plausibility of the specific scenarios described, the timelines involved, or arguing that focus on speculative long-term risks can distract from addressing more immediate, already-documented AI harms. Reasonable, well-informed people in the field disagree, and this disagreement is a defining feature of the topic rather than something to resolve toward one “correct” answer.

Bottom Line

AI existential risk refers to concern about the possibility that sufficiently advanced future AI systems could cause catastrophic and potentially irreversible harm to humanity, centered on the challenge of reliably aligning highly capable AI systems with human values. This is a distinct, more extreme concept than near-term AI harms, and it remains a genuinely disputed topic, with credible experts holding a range of views on its plausibility and urgency.

Go deeper

• Should someone in a non-technical field worry about AI replacing their job soon?