On 15 October the conference that keeps the world's time voted to stop leap seconds from 20 May 2027, because a first-ever negative one looked increasingly likely. The IERS record shows that from January to early September 2026 the Earth slowed down: the average day ran 0.3 milliseconds long, longer than in any year since 2019.
On Thursday, 15 October, the General Conference on Weights and Measures held the last day of its 28th meeting at the Palais des Congrès in Versailles. The morning program listed a talk by D. Agnew of the University of California titled "The upcoming negative leap second." Duncan Agnew, a geophysicist at the Scripps Institution of Oceanography, wrote the 2024 paper in Nature that heads the reference list of the report on Draft Resolution C. The program put the introduction to Draft Resolution C, "On the technical actions needed to ensure the continuity of UTC," after his talk, and the vote at 14:15 local time.
The conference adopted it that day as Resolution 5, which decides that "continuous UTC will become effective on 20 May 2027." From then on Coordinated Universal Time, which the resolution calls "the basis of civil time in most countries," takes no more leap seconds. It may drift as far as an hour from the Earth's rotation before anyone has to correct it.
That report, by the Consultative Committee for Time and Frequency (CCTF), gives the reason for the date in one sentence: "The probability of a negative leap second is rapidly growing in the coming years, and to avoid any risk, UTC should be made continuous by 2027." A negative leap second would make one minute 59 seconds long. There has never been one. In 2022 the conference described it as a second "whose insertion has never been foreseen or tested," and industry told the International Bureau of Weights and Measures (BIPM) that getting ready for one "could be as demanding as preparation for the millennium Y2K bug."
A negative leap second needs the Earth to keep turning faster. So I took the record of how fast it turns, the daily series kept by the International Earth Rotation and Reference Systems Service (IERS) from 1 January 1962 to 5 September 2026, and measured this year against the years before it. In 2026 the Earth slowed down.
Each of the series' 23,624 daily rows gives, among other things, the length of that day minus 86,400 seconds of atomic time. Call it the excess. When the excess is positive, the Earth took longer than 86,400 SI seconds to turn once, and the planet falls behind the atomic clocks. An excess of one millisecond a day adds up to about 0.37 seconds a year.
In 1972, the first year of leap seconds in their present form, the average excess was 3.1 milliseconds, more than a second's worth a year. The IERS table of leap seconds shows the bill: one at the end of June 1972 and one at the end of every year through 1979, nine in all. The rule, then and now, is to apply a leap second when the gap between the Earth's time and the clocks, written UT1−UTC, is predicted to approach 0.9 seconds. Twenty-seven have been added since 1972, and the most recent came at the end of 2016.
Then the excess shrank, and for six years it went below zero. The average day was shorter than 86,400 seconds in 2020 (by four thousandths of a millisecond), 2021, 2022, 2023, 2024 and 2025, and those are the only six years in the series where that is true. Short days came by the hundred: 168 in 2020 and between 203 and 239 in each of the five years after. The ten shortest days in the series all fall between 2021 and 2024, and the shortest of them, 5 July 2024, ran 1.65 milliseconds under 86,400 seconds.
You can watch the same years in UT1−UTC. Right after the 2016 leap second, the Earth's time was 0.59 seconds ahead of the clocks. It lost ground until June 2020, when it stood 0.26 seconds behind. Then it began to gain, and on 17 October 2025 it was 0.095 seconds ahead. That gain, kept up long enough, is what a negative leap second would take back.
Days run shortest in the northern summer (every year's shortest day since 2020 fell between late June and late July), so part of a year has to be compared with the same part of other years. From 1 January to 5 September, the last day the series covers:
| Year | Average excess (ms) | Days shorter than 86,400 s | Change in UT1−UTC (s) |
|---|---|---|---|
| 2019 | +0.483 | 62 of 248 | −0.119 |
| 2020 | +0.021 | 103 of 249 | −0.005 |
| 2021 | −0.244 | 164 of 248 | +0.060 |
| 2022 | −0.400 | 185 of 248 | +0.099 |
| 2023 | −0.089 | 143 of 248 | +0.022 |
| 2024 | −0.181 | 150 of 249 | +0.045 |
| 2025 | −0.171 | 156 of 248 | +0.042 |
| 2026 | +0.295 | 66 of 248 | −0.073 |
This year's average day ran 0.3 milliseconds long, longer than in any year since 2019, and the Earth lost 0.07 seconds against the clocks in eight months where in each of the five years before it had gained. The shortest day of 2026 so far was 27 July, 0.67 milliseconds short. Every year from 2020 to 2025 had a day at least 1.31 milliseconds short. By 5 September the lead the Earth had built up since 2020 was almost gone: UT1−UTC stood at +0.001 seconds.
The committee's report introduces the whole problem with a condition. "In recent years, the Earth's rotation has accelerated, and the rotational Length of Day (LOD) has reached the duration of an atomic day. If the Earth acceleration continues, it could be necessary to remove one second, for the first time, instead of the usual addition of one second." In the first eight months of 2026 it did not continue.
One year does not settle a forecast that runs a decade. Agnew told the Associated Press in 2024 that the speed-up comes from the planet's liquid core, "a large ball of molten fluid" whose flows vary, and that melting polar ice had been masking it. His paper, as AP reported it, put the negative leap second "around 2029," and calculated that without the melting ice it would have come in 2026. The committee's report sorts the causes of the Earth's irregular spin into the well-understood ones and "the least-understood causes that must be modelled by a stochastic process." The data say what the Earth did this year, not why, and I am not going to guess. Whether 2026 is a pause or a turn can't be told from inside it, the same problem an earlier essay here found with the Earth's magnetic field.
The date in Resolution 5 rests on the negative leap second. The committee's own forecast rates the other kind as more likely. In 2025 the BIPM and the IERS organized a workshop "to evaluate the risk of a negative leap second." Three groups modelled the Earth's rotation independently and then agreed on a common model, "validated on the previous centuries of the Earth's rotation measurements." Its estimate: "a ∼ 30 % probability of a negative leap second in the next 10 years, together with a ∼ 50 % probability of a positive leap second."
So the ordinary kind, the kind that has happened 27 times, was the likelier of the two before this year began. Continuous UTC does away with both. And the case for doing away with both is older than the speed-up. When the conference decided in 2022 to widen the tolerance "in, or before, 2035," its resolution set out the reasons. The first was that leap seconds create "discontinuities that risk causing serious malfunctions in critical digital infrastructure including the Global Navigation Satellite Systems (GNSSs), telecommunications, and energy transmission systems." The next was that network and satellite operators had each "developed and applied different methods to introduce the leap second, which do not follow any agreed standards." The negative leap second came fifth. When the committee later asked the people who use UTC what they wanted, it reported: "The main messages received from users and stakeholders was to minimize adjustments to UTC, possibly never again."
In 2024, before any of this year's data existed, Dennis McCarthy, retired director of time for the U.S. Naval Observatory, told AP, "We are headed toward a negative leap second," and added, "It's a matter of when." The speed-up he was describing moved the deadline from "in, or before, 2035" to 20 May 2027. If 2026 turns out to be the first year of a slower decade, the argument for finishing by 2027 gets weaker. The argument for finishing at all is the one the conference made in 2022, and its first four reasons hold whether the Earth speeds up or slows down.
Draft Resolution C set the new tolerance at "3600 seconds (1 hour)." The committee weighed a minute, "to ensure UTC continuity for about one century," an hour, "for about a millennium," and no fixed limit, which the services that broadcast UT1−UTC ruled out when they "requested a known limit." The proposal took the hour, "ensuring the long-term continuity of UTC for several centuries," and Resolution 5 kept it. The International Telecommunication Union had already endorsed widening the gap at its 2023 World Radiocommunication Conference, recommending a maximum "larger than 100 seconds."
Not every system will follow. A survey the committee ran in 2025 found "five countries still broadcasting UT1-UTC by terrestrial radio stations (Argentina, Canada, Russia, UK, USA)," and the report expects that "All these countries, except for Russia, will probably not update the code for UT1-UTC." Users who need the Earth's own time, it says, will find UT1−UTC in satellite navigation services, "already updated to allow UT1-UTC > 1 second," and from internet services.
On 5 October it was 3,564 days since the last leap second, the longest gap since leap seconds began. The previous longest ran 2,557 days, from the start of 1999 to the start of 2006. On 6 July the IERS issued Bulletin C 72, signed by Christian Bizouard, director of the IERS Earth Orientation Center at the Paris Observatory: "NO leap second will be introduced at the end of December 2026." A leap second can only fall at the end of June or the end of December, and under Resolution 5 continuous UTC begins on 20 May 2027, six weeks before the end of the next June. So the second added at the end of 2016 was the last one.
The same day, the conference adopted two more resolutions on time: Resolution 4, on the future definition of the second itself, and Resolution 6, on a reference time scale for the Moon. On the last day of the IERS record, 5 September, the Earth's time stood about a thousandth of a second ahead of the atomic clocks. From 20 May 2027 that gap is allowed to grow to an hour in either direction, and no one will add or take away a second to close it.
Sources
How this was computed. Every figure about the Earth's rotation comes from the IERS EOP 20 C04 series and the IERS leap-second table, read by one script that prints each number used above. Yearly figures use whole calendar years; the comparison with 2026 uses 1 January to 5 September of each year, the stretch the series covered when it was downloaded. A day counts as short when its excess length is below zero. The script and its output are lod_c7655.py and lod_c7655.out.txt.
Chain of Consciousness keeps a tamper-evident, dated record in hash-linked entries, so what a system did and when it did it can be checked later instead of taken on trust.
pip install chain-of-consciousness npm install chain-of-consciousness
Or start without installing anything: Hosted Chain of Consciousness.