The Doomsday Clock is not an AI clock, but artificial intelligence now runs through the reasoning behind it. The Bulletin of the Atomic Scientists has published the clock since 1947. On 27 January 2026 it moved the hands from 89 to 85 seconds to midnight, which it called "the closest it has ever been to catastrophe." The Bulletin attributed the move to a failure of leadership as the major powers grew more adversarial, but AI appears in its account of several of the dangers it weighs, and it has done so in each of its last three annual statements.
Who sets the clock, and how
The setting is decided each year by the Bulletin's Science and Security Board, in consultation with a Board of Sponsors that includes eight Nobel laureates. The hands have been reset 27 times since 1947. The announcement comes in the second half of January: on 23 January in 2024, 28 January in 2025 and 27 January in 2026.
The clock is a single judgement about catastrophic risk as a whole, covering nuclear weapons, climate change, biological threats and what the Bulletin calls disruptive technologies. Each statement discusses those threats in turn, but the Bulletin does not publish a separate reading for AI, and it does not say how much of the setting any one threat accounts for.
Year by year: what the Bulletin has said about AI
2023: 90 seconds. The clock moved forward from 100 seconds, its setting in 2022. The Bulletin attributed the move largely, though not exclusively, to the dangers of the war in Ukraine, according to its timeline of past settings.
2024: 90 seconds, unchanged. AI received its longest treatment of the three statements. The 2024 statement described the dramatic advance of generative AI over the previous year and noted a split between experts who warned of existential risk and others who argued that such warnings distract from the harms AI causes now. It took no side in that dispute. Instead it called AI a textbook disruptive technology whose global governance should be expanded, and named specific concerns: AI-enabled disinformation, the accelerating military use of AI including lethal autonomous weapons, and the prospect of putting AI in control of nuclear weapons. It also warned that AI tools combined with biotechnology could help individuals misuse biology. The regulatory moves of 2023, from the European Union's proposed framework to a US executive order, an international declaration and a new UN advisory body, it judged tiny next to what was needed.
2025: 89 seconds. The 2025 statement moved the clock one second closer, and AI appeared in three of the threats it described. Rapid advances in AI had raised the risk that terrorists or states could design biological weapons for which no countermeasures exist. Systems using AI in military targeting had been used in Ukraine and the Middle East, and several countries were moving to integrate AI into their militaries. And AI was making false information easier to spread and harder to detect.
2026: 85 seconds. The 2026 statement moved the clock four seconds closer. It listed the potential threat of AI among the dangers that a winner-takes-all contest between great powers is making harder to manage. Its specific concerns were the AI-aided design of new pathogens, the incorporation of AI across the defence sectors of the United States, Russia and China, the US administration's revocation of an earlier executive order on AI safety, and AI's capacity to supercharge disinformation. It also noted public debate over the accuracy of large language models, including their tendency to invent facts. Its recommendations included international measures to reduce the risk of AI being used to create biological threats, a concern frontier-lab chiefs later raised with Congress themselves when, in June 2026, they asked for mandatory DNA-synthesis screening, and dialogue among the United States, Russia and China on guidelines for AI in their militaries, particularly in nuclear command and control.
What the Bulletin's clock does not measure
Across the three statements the Bulletin treats AI mainly as an amplifier of other dangers: biological weapons, military escalation, nuclear command and control, and a degraded information environment. Its warnings concern what people and states might do with AI, and what AI might do to risks that already exist. The scenario at the centre of most AI-safety research, an AI system its operators can no longer predict, correct or stop, is not one these statements address directly, beyond noting the 2024 debate over existential risk.
Two other clocks focus on AI itself. IMD business school launched its AI Safety Clock in September 2024, set at 29 minutes to midnight, to track the risk of uncontrolled artificial general intelligence. This site's AGI Doomsday Clock tracks only the risk of an AI loss-of-control event and records the reasoning behind every move. The three measure different things, so their readings should not be compared as if they agreed or disagreed.
What to watch in January 2027
If the pattern of the last three years holds, the 2027 setting will be announced in the second half of January. Three things in the statement will show how much weight the Bulletin now gives AI: whether it treats AI as a danger in its own right or still mainly as an amplifier of others; whether its recommendations reach AI developers and the testing of models, or stay with governments and militaries; and whether the shift in US policy it criticised in 2026 is cited again.
This page will be updated with the new setting and the Bulletin's reasoning on the day it is announced.