Humanity is not merely approaching a precipice; it is sprinting toward one. Artificial intelligence promises extraordinary gains, but its rapid development may also pose an existential threat. Governments and companies around the world are pursuing AI for different reasons, with different tolerances for risk and remarkably little agreement about where the boundaries should lie.

The potential benefits are substantial. AI could improve healthcare and public safety, increase workplace productivity, make power grids more efficient, strengthen agricultural production, and accelerate the discovery of new materials. Governments can use it to streamline public services and bolster national security. They can also deploy it for surveillance, cyberwarfare, and military force multiplication. The same systems capable of improving daily life can be adapted to undermine it.

In the U.S., the pursuit of profit has too often eclipsed the demand for operational integrity. Innovation is treated as an end in itself, while Washington has largely declined to impose meaningful limits on the industry. The European Union has moved more aggressively, establishing guardrails intended to protect privacy, improve transparency, and ensure a measure of public benefit. China, meanwhile, is determined to reduce its technological dependence on the U.S. and other countries while using AI both to deliver services and to expand the state’s surveillance capacity.

The greatest danger comes from companies that place speed and market dominance above safety. Developers of frontier systems have acknowledged that they do not fully understand how their most advanced models arrive at certain conclusions, much less how those systems will behave in every circumstance. Yet they continue to release increasingly capable, increasingly autonomous tools to paying customers. Soon, AI agents may help design more powerful successors with far less human involvement. Washington’s response has so far been hesitant and fragmented. By the time policymakers grasp the scale of the risk, the most consequential decisions may already have been made.

The vulnerability is magnified by the way modern society is built. Water and energy systems, satellites, telecommunications networks, logistics and transportation, healthcare, financial services, cryptocurrency markets, and e-commerce are all connected to the Internet. Municipal, state, and federal agencies depend on digital networks, as do militaries, even when their most sensitive systems are isolated. Advanced AI agents could help attackers penetrate these sectors, disrupt essential services, and search for weaknesses faster than human defenders can repair them. Autonomous laboratories, meanwhile, raise a different nightmare: AI systems could make it easier to design dangerous biological agents.

Quantum computing could make the problem even more severe. Some experts warn that sufficiently powerful quantum machines may eventually defeat widely used forms of encryption, potentially exposing data collected years earlier. Post-quantum cryptography is being developed, but the transition will be uneven. Combined with advanced AI agents, quantum capabilities could give rogue systems, hostile states, or non-state cyberterrorists tools that today’s security architecture was never designed to withstand.

Even if AI does not produce a sudden catastrophe, it could destabilize societies more gradually. AI systems and robotics may replace large numbers of workers over the next five to ten years. If people lose their incomes, they cannot buy goods and services or pay taxes at the same level. Businesses weaken, government revenues decline, and social protections come under pressure. When economic displacement threatens access to food, housing, and basic security, civil unrest becomes less a remote possibility than a predictable political response.

Avoiding that future will require urgent action across national borders. Operators of critical infrastructure need contingency plans that allow essential systems to function offline, however difficult and expensive that may be. Governments must cooperate on enforceable rules for advanced AI rather than relying on voluntary promises from the companies building it. They must also prepare for mass labor displacement, whether through public employment guarantees, universal basic income, stronger social insurance, or some combination of the three.

The exponential pace of technological change makes it impossible to know precisely when a crisis will arrive. It may come sooner than expected, or unfold in ways that are harder to recognize: a cascade of cyberattacks, a hollowed-out labor market, an autonomous system behaving beyond its creators’ control. Some people remain unaware of the danger; others find it easier to dismiss. Either way, the central task is becoming harder to avoid. If humanity is to survive the technologies it is creating, it must first confront the human incentives driving them.