Who Can Start a Nuclear War? Inside U.S. Launch Authority and Reform

Erin D. Dumbacher is the Stanton nuclear security senior fellow at the Council on Foreign Relations.

The United States has the power and a process to respond when it is under nuclear attack, but only the president can decide if and when to launch the world’s most devastating weapons. Nuclear deterrence hinges on the notion that the U.S. military would retaliate as soon as the president gives a legal order. Past reforms—and even the policy that the president has sole authority to order a launch—have sought to ensure rogue military commanders or mistakes in missile silos, on submarines, or aboard bomber aircraft cannot launch a nuclear weapon without authorization. Today, reform options focus not on an unauthorized launch but on preventing an unwise first strike. 

The U.S. military is always ready to launch promptly if it or the intelligence community learns of an incoming attack and the president authorizes a response. Threats have evolved and technology has advanced, yet this policy has been consistent since the Cold War. The system is designed to prioritize speed when implementing the president’s decision. Military service members must carry out all legal orders to launch.

Some experts argue that it is the singular command, plus the option to launch when there is a warning of an incoming assault, that allows the president minutes to decide on a course of action and therefore most risks miscalculation. The United States has survivable nuclear forces, proponents of the current posture counter, and there is no requirement that the president direct the launch of nuclear weapons if systems indicate one or more nuclear delivery systems are en route to the United States. U.S. forces can withstand an initial attack; presidents need not rely on “launch-under-attack,” the Biden administration confirmed in its 2022 policies. The Trump administration has the option to reiterate this approach or to reform it through its own forthcoming National Defense Strategy.

Launching a Nuclear First Strike

Two U.S. presidents—Ronald Reagan and Joe Biden—have now affirmed, along with other permanent members of the UN Security Council including Russian leaders, that a nuclear war “cannot be won and should never be fought.” At the same time, U.S. declaratory policy and military practices allow the United States to use nuclear weapons first. Some outside of government have suggested reforms to shift the country to a “no first use” policy. China officially maintains a similar policy, although they are also expanding their nuclear arsenal substantially—likely reaching one thousand weapons by 2030. 

A president is not required to seek congressional approval to use nuclear weapons and may choose to use them before an adversary does. Many Americans—61 percent according to recent polling—are uncomfortable with the president having this sole authority to authorize use. Some members of Congress have introduced legislation to limit the president’s ability to launch a first strike (or to limit the use of federal funds to support any first strikes), though few have signed on to the legislation, and it seems unlikely to move forward given the current politics. 

The Obama and Biden administrations both investigated ways to reduce the role of nuclear weapons in U.S. strategy and to clarify that the “sole” purpose of having them was to deter strategic attack. They aimed to limit the circumstances under which the United States would respond with nuclear weapons. The Biden administration’s strategy [PDF] clarified that the role of U.S. nuclear weapons was to respond to strategic attacks and to “achieve U.S. objectives if deterrence fails.” 

Restraint mechanisms to avert miscalculation remain a promising area for reform, reducing the possibility of a first use of a nuclear weapon based on inaccurate or mistaken information and building guardrails around a future president’s sole authority to launch a first strike. A miscalculated launch is arguably a higher risk today than massive strategic attack from other nuclear powers. Shifting to a sole purpose policy, however, competes with nonproliferation aims: ally and partner states without their own nuclear weapons, but who are under the U.S. nuclear “umbrella,” would like to see continued commitments from the United States to use nuclear weapons against adversaries who inflict extreme damage. 

Nuclear Retaliation and Risk of Mistakes

The logic of nuclear deterrence and U.S. policy holds that if an adversary strikes the United States with nuclear weapons, the United States will retaliate with nuclear weapons. Early on in the Cold War, strategists forecast not single strikes or dozens of missiles, but dramatic, vast salvos. These would arrive without warning and with enough nuclear weapons to decimate entire states and regions with radiological effects. The policy’s purpose is to deter or persuade the adversary not to launch a nuclear weapon in the first place. 

U.S. nuclear deterrence policy likewise still reflects these fears, resting on a commitment to retaliate against any nuclear strike and maintain a “launch on warning” posture, meaning U.S. weapons are ready to launch immediately upon detection of an incoming threat. Today, most analysts agree the greater risk is miscalculation or escalation during conventional warfare or a crisis that could lead to nuclear weapons use, including on a battlefield. The United States said it would use a series of conventional weapons and forces to counter Russian President Vladimir Putin’s October 2022 threat to use a nuclear weapon on a Ukraine battlefield. So far, Putin has found non-nuclear ways to escalate his war in Ukraine. 

Few reforms to U.S. policy would reduce the primacy of the U.S. president and his or her ability to respond to a launch warning against the United States. However, current policy allows the president to “ride out” an attack—letting the United States absorb a nuclear strike that would have devastating human and environmental consequences but would not eliminate the military’s ability to strike back with remaining forces. This allows more time for the president and advisors to understand events and reduces the possibility of launching retaliatory strikes based on a false alarm. Of course any nuclear detonation on U.S. territory would come at a catastrophic cost.

Other proposals would try to find more “decision time” for the president. Today, in only a matter of minutes, the intelligence community would identify an incoming attack, try to validate it with multiple sources, and communicate information about it through the national command structure. Given an intercontinental ballistic missile (ICBM) from Russia to the continental United States has a flight time of roughly thirty minutes, the president has only a few moments to consider this information and decide how to respond. One proposal would have the president “decide under attack”—learn about an adversary assault, make a choice to retaliate, but order the military to delay execution until a designated, later hour. Yet there are deterrence and assurance problems with this approach that make military adoption unlikely. 

Past reforms have sought to limit the risk of unauthorized launch or accidents. Since the 1960s, specialized systems called permissive action links have prevented the unauthorized use of U.S. nuclear weapons. Accidents in Spain and Greenland led Congress to constrain the military’s exercises and training: the Air Force may no longer fly training missions with real nuclear weapons. And President Truman put the practice of sole authority itself into effect to limit the scope of his generals at the end of World War II.

U.S. policy now also prevents handing this decision off to machines or artificial intelligence (AI), which some fear could lead to catastrophe. This safeguard, echoed in an agreement between Biden and Chinese President Xi Jinping in 2024, ensures that only humans can make the decision to use nuclear weapons. (I was fortunate to work on the specifics of that commitment while I was in government.) The president remains the final check on both the military and intelligence community, as well as all of their supporting machines and equipment. 

Adjusting to Technological Changes

Emerging technologies like hypersonics or other maneuverable weapons pose unique challenges to U.S. nuclear posture. Adversaries who threaten to arm their hypersonic vehicles with nuclear weapons are raising the stakes and potentially undermining stability. (U.S. hypersonic capabilities are, as a policy matter, for “conventional” or non-nuclear use only.) These new technologies can introduce uncertainty in an already complex geopolitical environment. 

Most analysts agree that any use of a nuclear weapon in the United States, against a U.S. adversary, or among other states with such weapons—like India and Pakistan—would have disastrous humanitarian and political consequences. Scholars believe this has created a “nuclear taboo”—a sense that even though nine countries have the weapons, detonating them again is so stigmatized that leaders do not feel at liberty to use them. President Trump said as much in his speech to the UN General Assembly this year, stating, “we just can’t ever use them. If we use them, the world might literally come to an end.” 

For good reason, leaders are concerned about the destructive effects of a nuclear explosion. Today’s nuclear weapons are larger than the first so-called Little Boy and Fat Man that the United States dropped on Japan during World War II. The number of nuclear weapons, especially in the U.S., Russian, and Chinese stockpiles, get significant attention in policy discussions, but the ability for an ICBM to contain multiple nuclear warheads—and for those warheads to have even more decimating effects than the first nuclear weapons, if detonated—is sometimes overlooked. The bombs that devastated Hiroshima and Nagasaki were relatively primitive 16 and 21 kiloton (kt) bombs. The nuclear warheads inside U.S. weapons today can be as small as 8kt, but can be upwards of 475kt. U.S. submarine-launched ballistic missiles could yield 100kt; ICBMs in the arsenal would yield hundreds of kt more. 

It would be easy to assume that AI is altering nuclear weapons policy and deterrence like it is changing economies. But that’s not necessarily the case. In the modernization of U.S. nuclear weapons currently underway, most upgrades [PDF] involve shifting from analog to digital while not necessarily adding in machine learning or advanced automation. The exceptions are likely in nuclear planning systems, intelligence data integration, or early warning capabilities.  

The key AI risk to nuclear systems is in feeding inaccurate information into the intelligence or command and control system. AI doesn’t need to play a role in some of the world’s most important computer systems just because it can. The Pentagon’s testing and evaluation processes—backed by the military services and their best engineers and scientists—should only insert advanced systems where they can deliver important benefits for stability and reassurance. Generating high-quality data to train an early warning AI model is a notable challenge. What’s more, deep fakes in the news or on social media could proliferate falsehoods that distort leaders’ perceptions during a crisis. 

Chance for Reform

Cold War threats led to the United States’ current approach to nuclear weapons policies, guidance, and capabilities. Today’s risks—from AI to an expanding Chinese arsenal—will guide U.S. policy in the coming decades. Substantive reforms, like checks on the first use of a nuclear weapon, could help bring U.S. nuclear weapons in line with the stated ambitions of presidents and leaders of nuclear states around the world: a nuclear war cannot be won and, therefore, should never be fought.

This work represents the views and opinions solely of the author. The Council on Foreign Relations is an independent, nonpartisan membership organization, think tank, and publisher, and takes no institutional positions on matters of policy.