A new assessment published in The Lancet identifies malicious use of artificial intelligence and nuclear conflict as threats with the potential to cause human extinction. Looking ahead to 2100, the Commission examines 17 interconnected risks to health and survival, spanning technology, conflict, environmental change and social conditions. Its central message is that decisions made today can substantially reduce future harm.

The assessment brings together expertise from 35 leaders in health, economics, security, governance and technology, drawing on analysis covering 204 countries and territories. It is co-chaired by Christopher Murray, director of the Institute for Health Metrics and Evaluation at the University of Washington, and Natalia Kanem, former executive director of the United Nations Population Fund.

The distinction between a risk assessment and a prediction matters. The Commission explores possible outcomes and opportunities for prevention; it does not forecast inevitable extinction or provide a fixed ranking of policy priorities. Its findings are intended to evolve as evidence changes.

Alongside AI misuse and nuclear conflict, the assessment considers threats including climate change, antimicrobial resistance, pandemics, air pollution, childhood malnutrition and income inequality. Some already cause substantial harm, while others could have catastrophic consequences if they materialise. Addressing them requires cooperation across sectors because their effects can reinforce one another.

To compare these different risks, researchers modelled plausible best- and worst-case futures using disability-adjusted life years, or DALYs—a measure of healthy life lost through illness, disability or premature death. The gap between scenarios helps identify where intervention could prevent the greatest losses. The framework includes a threshold of at least one billion DALYs by 2100, alongside sufficient evidence for modelling and scope for policy action.

For technology and business leaders, our reading is that AI risk deserves attention at the level of the entire system in which it operates. Evaluating a model’s performance is one part of that work. Understanding its access, authority, dependencies and potential consequences is equally relevant when introducing AI into important workflows.

This creates practical questions for organisations building or adopting AI. What information can a system access? Which actions can it initiate? Where is human approval required? How can teams detect unexpected behaviour, stop an operation and restore normal service?

These are engineering implications we draw from the broader discussion, rather than specific recommendations from the Commission. They connect long-term concerns about technology to decisions teams can make during architecture, implementation, testing and release.

The Commission’s own agenda focuses on building health systems prepared for emerging threats, increasing investment in breakthrough innovation and regularly reassessing risks. It calls for sustained preparedness and coordination across health, finance, climate, defence and security.

At Control F5 Software, we see responsible AI adoption as a continuing engineering responsibility. Clear permissions, meaningful testing, traceable actions and accountable human oversight help teams introduce useful capabilities while retaining control over the systems their operations depend on.

Source