Automated Nuclear Launch Systems Could Strengthen Coercion

Automated Nuclear Launch Systems Could Strengthen Coercion

As geopolitical tensions escalate and arms control frameworks erode, a troubling possibility has emerged: nuclear-armed states could use artificial intelligence to automate launch decisions, making their threats more credible and coercive.

New experimental research demonstrates that when adversaries believe nuclear retaliation has been delegated to automated systems, they perceive threats as significantly more believable — and become more willing to capitulate. The findings, first reported by War on the Rocks, suggest that automation could fundamentally alter nuclear deterrence dynamics in ways that increase rather than reduce risk.

Why it matters

This isn't theoretical. Russia already deployed a semi-autonomous system called Dead Hand during the Cold War, designed to ensure retaliation even if leadership was eliminated. Today's AI capabilities make full automation technically feasible, and the strategic incentives for states facing conventional disadvantages or decapitation fears are real. Understanding how automation affects threat credibility is essential for policymakers navigating an era where nine nations possess nuclear weapons and traditional arms control agreements have collapsed.

The Credibility Problem

Nuclear coercion typically fails for two reasons. First, mutually assured destruction makes threats irrational when adversaries possess secure second-strike capabilities — carrying out an attack invites devastating retaliation. Second, the nuclear taboo imposes political and social costs that undermine threat credibility.

Automation changes this calculus. By programming launch decisions into computer systems, leaders can credibly tie their own hands. Unlike humans, algorithms don't calculate political costs or fear retaliation. Once activated, an automated system becomes what researchers call "irrevocable" — removing the human judgment that typically prevents nuclear use.

Testing Automated Threats

Researchers surveyed over 100 UK Members of Parliament and more than 1,000 British citizens, presenting a scenario where Russia invades Estonia in 2030 and threatens nuclear strikes if NATO intervenes. Half the respondents were told Russia's threat relied on automated launch systems triggered by sensors detecting NATO involvement.

The results were stark. Members of the UK public were 13 percentage points more likely to believe Russia would follow through when automation was involved. UK parliamentarians were nine percentage points more willing to abandon Estonia — a NATO ally protected under Article 5 — when facing automated threats.

These effects persisted despite the UK's own nuclear arsenal and despite some inconsistency in how parliamentarians assessed threat credibility. The mere possibility that an adversary had removed human control created enough uncertainty to shift strategic calculations.

Real-World Risks

The dangers extend beyond hypothetical scenarios. During the Cold War, false alarms from Soviet early-warning systems nearly triggered nuclear retaliation — disasters averted only by human judgment. Modern systems face similar vulnerabilities, from software failures like Boeing's 737 MAX crashes to hacking risks that could enable malicious actors to trigger launches.

Yet states may still pursue automation. Countries fearing decapitation strikes could automate to ensure retaliation capability. The accelerating speed of AI-enabled conventional warfare could pressure states to automate nuclear command to prevent rapid defeat. And states losing on the battlefield might gamble on automated nuclear coercion as a last resort.

Policy Implications

The research suggests several responses. Nuclear-armed states should strengthen second-strike capabilities to reduce incentives for adversaries to pursue automation. Governments could impose costs — such as economic sanctions — on states developing automated launch systems. And while verification would be extraordinarily difficult, a multilateral treaty prohibiting nuclear automation could raise the political costs of noncompliance beyond the informal agreements that currently exist between Washington and Beijing.

The findings were reported by Joshua Schwartz and Michael Horowitz in War on the Rocks, based on their survey research conducted in 2026.