The next few decades are the most dangerous in human history

The risks we face are not a list. They are a system.

This is the single most important thing to understand about the danger ahead, and almost every analysis gets it wrong. The standard approach treats existential risks as independent line items: nuclear war, artificial intelligence, engineered pandemics, climate change. Each gets its own chapter, its own experts, its own policy proposals. But the risks do not exist in isolation. They interact, compound, and amplify each other in ways that make the whole dramatically more dangerous than the sum of its parts.

AI acceleration compresses the timeline for every other risk. Advances in artificial intelligence are making it easier to design biological weapons, faster to identify vulnerabilities in critical infrastructure, and harder for institutions to keep pace with the rate of change. Economic disruption from AI, which may displace enormous numbers of jobs in the coming years, will generate political instability that reduces the capacity of governments to coordinate on climate, biosecurity, and AI governance itself, precisely when the need for coordination is greatest.

Climate change destabilizes everything else. It is probably not an extinction risk on its own, but it is a civilizational stress multiplier. Climate refugees create political pressure. Agricultural disruption increases competition for resources. Both fuel the nationalist backlash that is already undermining the international cooperation needed to address every other challenge on this list. Climate doesn't need to end civilization directly. It just needs to make us too fractured and distracted to deal with the things that can.

Biotechnology is being democratized by AI. The knowledge required to engineer dangerous pathogens once demanded years of specialized training. AI is compressing that barrier. Current models may already be doubling or tripling the probability of successful bioweapon creation by non-experts. This risk grows with every improvement in AI capability, on timescales of months.

Nuclear weapons have not gone away. The Cold War standoff has been replaced by a multipolar nuclear landscape that is in some ways more dangerous because it is less predictable. We survived the nuclear age so far through a combination of institutional safeguards and sheer luck. The luck is not a strategy. The safeguards are eroding.

These risks form feedback loops. AI makes biotech more dangerous. Biotech crises increase pressure on governments. Government dysfunction reduces the capacity for AI governance. Climate disruption destabilizes the political landscape, weakening international cooperation on everything. Pull on any thread and the whole web moves.

Why we know we can't handle this

COVID told us everything we need to know about our species' capacity to coordinate under pressure. And the lesson is worse than most people realize.

A novel respiratory virus emerged, almost certainly leaked from a gain-of-function research laboratory. It spread globally. The death rate was roughly 1%. Governments issued contradictory guidance for months. The scientific establishment insisted the virus could not have come from a lab, then quietly reversed itself. The evidence behind key public health interventions like masking remained contested throughout. Vaccines were developed at unprecedented speed, a genuine triumph, but their efficacy was overstated in ways that eroded trust when breakthrough infections became common. The question of whether vaccination provided meaningful additional benefit to those who had already been infected and recovered was never honestly addressed. The information ecosystem became so polluted that ordinary people had no reliable way to determine what was true. Political leaders weaponized the crisis. International coordination was dismal. More than six million people died.

This was not a hard test. A pandemic with a 1% death rate, caused by a natural pathogen (or at minimum an accidental release rather than a malicious one), is about as manageable as global crises get. The threat was immediate, visible, and universal. Everyone was affected. Everyone knew they were affected. The incentive to coordinate was as strong and as aligned as it will ever be.

And we failed. Not partially. Comprehensively.

Now consider what this implies about our ability to manage the risks that actually matter.

A pandemic has immediate, visible, universal consequences. The bodies pile up. The hospitals overflow. Everyone can see the problem. AI is the opposite. The immediate consequences of AI development are benefits: enormous productivity gains, competitive advantages, profits. The potential negative externalities, displacement of hundreds of millions of workers, concentration of power in the hands of a few companies or governments, the risk of misaligned superintelligence, are diffuse, delayed, uncertain, and easy to dismiss until they arrive. If we cannot coordinate when everyone is visibly dying from the same threat, what reason is there to believe we can coordinate when the immediate incentives all point toward moving faster?

The gain-of-function point cuts even deeper. We could not even coordinate to keep a virus inside a laboratory, which is among the simplest coordination problems imaginable. Contain the dangerous thing. Don't let it out. Or don't pursue gain of function research to begin with. We failed at that. Governing the development of superintelligent AI across competing nations, competing companies, and competing ideologies, with trillions of dollars and possibly the future of civilization at stake, is a coordination problem orders of magnitude more complex. There is no honest reason to believe our current institutions are remotely capable of it.

The trap

And this is where the situation becomes genuinely frightening, because the structure of the problem makes voluntary restraint nearly impossible.

The potential rewards of being first to develop transformative AI are so vast, not merely financial but strategic, possibly civilizational, that no major actor can afford to slow down alone. America cannot let China get there first. China cannot let America get there first. Within the American ecosystem, each company believes that it is better for the world if they, rather than their competitors, lead the development. Each may be right about their relative merits. That is not the point. The point is that the race structure ensures that everyone moves as fast as possible, because the cost of falling behind feels greater than the cost of moving too fast. Caution is punished. Speed is rewarded.

This dynamic extends beyond AI. In biotechnology, the same research that enables life-saving gene therapies lowers the barrier to engineered pathogens. In climate, nations that impose costly carbon restrictions on their industries lose competitive advantage to nations that don't. In every domain, the structure of competition creates a race to the bottom in which restraint is punished and recklessness is rewarded.

Dario Amodei, the CEO of Anthropic, describes AI as "so powerful, such a glittering prize, that it is very difficult for human civilization to impose any restraints on it at all." He runs one of the companies building it. He understands both the promise and the danger as well as anyone alive. And he is telling us, plainly, that the system he operates within may not be governable by the institutions we currently possess.

The gauntlet

Multiple existential-scale risks, arriving simultaneously, interacting in a web of amplifying feedback loops. A species that failed to coordinate on a 1% mortality pandemic and has no honest basis for believing it can coordinate on challenges orders of magnitude harder. A political economy that makes voluntary restraint structurally impossible. And a shrinking window of time.

The path through this gauntlet is narrow. But it exists.