Lancet Commission Names Malicious AI and Nuclear War Among Top Threats Through 2100

A new global commission has assessed 17 catastrophic threats to human health over the next 75 years. The Lancet published the assessment, which the Guardian covered. Two items top the list: malicious use of AI and nuclear conflict. According to the authors, either one could cause human extinction by 2100. The commission stresses that every one of the 17 threats has a "realistic path for action."

What made the list

The published excerpt gives only a general picture, but it is telling:

• Number of threats: 17, all classed as catastrophic for human health.
• Horizon: 75 years, through 2100.
• Most severe risks: malicious use of AI and nuclear conflict, both with the potential for extinction.
• Other items: cuts in development aid from wealthy countries, the climate crisis, child malnutrition.
• Conclusion for each risk: a realistic path for action exists.

The mix is unexpected. Child malnutrition and shrinking aid budgets don't look like end-of-the-world scenarios. But the commission measures threats to health, not how cinematic a catastrophe is. Slow, unglamorous processes, such as falling health funding in poor countries, can kill more people than a single dramatic event.

Why AI ended up next to nuclear weapons

The wording "malicious use" matters. It is not about machines rebelling, but about people using AI to cause harm. The Guardian's article does not disclose the commission's specific scenarios. Discussions of AI risk in the field usually cover several areas: help in building biological weapons, large-scale cyberattacks on critical infrastructure, including hospitals, mass disinformation and autonomous weapons systems. Which of these the commission considers a priority is something to check in the full report.

For the industry, this is a different kind of signal from the usual statements by AI labs. Here AI is assessed from a public health perspective and placed beside a threat humanity has lived with since 1945. That view sits poorly with presenting AI as a set of product features. It moves the conversation about risk out of the "model safety" category and into "global health threats," where different regulators, different budgets and different audiences operate.

What to do about it

First: don't mistake a headline for a study. The claim of "extinction by 2100" sounds dramatic, but the published excerpt gives no methodology, probabilities or assumptions. Before citing the assessment, find the full report and check how the commission calculated the risks and where it got the 75-year horizon.

Second: pay attention to the phrase "realistic path for action." For developers and companies, it means checking for malicious use: tests for assistance with biological and cyber attacks, limits on access to the most powerful models, and request logging. For regulators, it points to international agreements on AI alongside nuclear controls. For ordinary readers and investors, the takeaway is more modest: AI safety will increasingly appear in documents from organizations that don't work on technology. It is worth following such publications, because future norms and requirements grow out of them.