Once every four to six years, the General Conference on Weights and Measures meets in Versailles—the body that decides what a metre, a kilogram, and a second mean for the entire planet. Its 28th session will take place on 13–15 October 2026, and item sixteen on the agenda is Draft Resolution C—on the continuity of Coordinated Universal Time, UTC.
Continuity here is no metaphor. Since 1972, humanity has lived with leap seconds: when the gap between atomic clocks and Earth’s actual rotation approaches one second, the IERS announces an extra second, and one minute lasts sixty-one. Resolution C proposes to end the practice: let the discrepancy grow to a full hour and leave the second alone.
The timing is awkward. Over the past few years, Earth has spun faster, and for the first time in history we may need not to add a second, but to take one away. That is precisely what nobody knows how to do—and, judging by the documents, nobody wants to.
The same paper surfaced the other day in a piece about the Martian day: there, it appeared as an addendum to the story of how people on Mars have spent twenty years living by homemade clocks. The thread turned out to be longer than it looked, and it leads to how they are about to fix the clocks on Earth this October.
The clock on your phone does not show the same thing as the sky above your head, and the gap between them is as old as timekeeping itself. Atomic time (TAI) counts cesium oscillations and never loses its beat. Earth’s rotation is a temperamental thing: slowed by tidal friction from the Moon, sped up by currents in the liquid core, changing from decade to decade. As long as time could be kept only as accurately as a pendulum, the difference did not matter. Atomic clocks arrived in 1955, and it became clear that humanity now had two incompatible versions of time—astronomical and physical.
The compromise came in 1972: UTC keeps pace exactly with atomic time, but sometimes has a second added so it does not fall behind the sky. That gave rise to the rule that UTC must stay within 0.9 seconds of UT1—the “Earth angle.” Over more than forty years, a second has been added twenty-seven times, most recently on 31 December 2016. Since then, TAI − UTC has been frozen at 37 seconds and has not changed for ten years.
The oddity is that leap seconds cannot be scheduled in advance. The IERS inserts them, announcing the decision roughly half a year beforehand: what is unclear is not even “when,” but “by what rules.” The signal is sent for the end of June or December, but only because that is how history happened to settle it.
Throughout the second half of the 20th century, Earth was slowing down, and seconds were added more and more often. Then the little motor changed gears. The record for the shortest day, unbroken since atomic clocks were first switched on, began falling every year: on 5 July 2024, the day was 1.66 milliseconds shorter than standard, and in 2025 Earth was once again expected to make several runs at the record over the summer.
Why? Geophysicists have an answer, and it sounds like a joke. On one hand, the planet’s liquid core has been speeding up its rotation for roughly half a century. On the other, over that same time, melting polar ice has been shifting water from the poles to the equator—and Earth behaves like a figure skater spreading her arms: redistribute the mass, and the rotation slows. The two effects work against each other, and that bought humanity some time.
Duncan Agnew of the Scripps Institution of Oceanography calculated exactly how much. In a paper in Nature (March 2024), he showed that without the contribution from melting ice, a negative leap second would have been needed as early as 2026, while with it, the deadline moved three years to the right—the author himself calls it “somewhere around 2028.” The press immediately rewrote that as “2029,” and half the story lives in that discrepancy: nobody knows the date exactly because nobody knows what the core will do.
Another figure from the same paper sounds almost quiet, yet says more than all the rest: over fifty years, the day has become 0.0025 seconds shorter. This is not a shift in the calendar. It is an admission that the planet has stopped being the standard.
Forty years of adding seconds have passed relatively quietly, but not without casualties. On 30 June 2012, an extra second brought down the Amadeus booking system: more than 400 Qantas flights were delayed, airport staff switched to manual check-in, and Mozilla, Reddit, Gawker, LinkedIn, Foursquare, and Yelp all ran into Java-service failures at once. Amadeus explained itself dryly at the time—“a Linux bug caused by the leap second”—and that is perhaps the most precise description of the problem in its entire history.
The next second, on New Year’s 2017, exposed something scarier than a bug. Cloudflare’s code contained a spot that calculated a DNS resolver’s response time: the difference between two moments which, by design, was “zero at worst.” When time went back by a second, the difference became negative, smoothing stretched that minus across several measurements, and the random-number generator rand.Int63n received a negative number as input and honestly panicked. Fewer than 1% of HTTP requests were affected; it took one check and a restart to fix. But the wording of Cloudflare’s apology still serves as the epitaph for the entire practice: “programmers believe time cannot go backwards.”
That is where the fear palpable in BIPM documents comes from. Humanity took its knocks with the positive leap second, but somehow survived. A negative second has never happened: in IERS data since 1972, every entry adds a second and not one subtracts it. In 2022, Meta wrote outright that such an event had “never been tested at scale” and could have “devastating impacts” on schedulers and timers. The industry, meanwhile, has assessed preparation for a negative leap second as comparable to preparation for the year 2000 problem.
Engineers stopped waiting for a global solution long ago and acquired their own—and that, it seems, worries the keepers of standards more than the seconds themselves. To avoid a clock jump inside a cluster, major companies do not insert the second all at once, but “smear” it across the day, slowing their clocks ever so slightly. The practice is sound, but every outfit has its own version.
Since 2008, Google has smeared the second linearly over 24 hours, from noon to noon UTC, and recommends precisely that method to everyone: the deviation comes to about 11.6 ppm, less than the error of quartz oscillators and forty times lower than NTP’s maximum correction rate. The company used to stretch the second across 20 hours, then switched to a day to match Amazon. Meta does it differently: it smears it quadratically over 17 hours, starting at midnight UTC. There is no agreed standard for “smearing,” everyone has a different starting point, and the gap between smeared and honest UTC reaches half a second—and it is precisely this discord that the draft resolution calls a threat to national infrastructures.
The result is a paradox. Everyone needs a single time, but every major operator already lives by its own slightly adjusted version of UTC, and the further this goes, the more systems depend on their particular clocks. Continuity as an argument for abolishing leap seconds cuts both ways: it also means everyone could keep smearing the second their own way indefinitely.
Astronomers oppose abolition, and their position is not about convenience. In 2005, the Royal Astronomical Society issued a statement calling the proposal at the time “unprofessional” precisely because it solved one group’s problem by exporting it to everyone else. Telescopes and satellite-tracking stations need mean solar time itself in order to point accurately; UTC with leap seconds served as their free approximation. Remove the second, and you take away their link between civil time and the sky.
There is also this, rather bitterly: the statement called the argument that leap seconds threaten navigation systems a distraction. GPS is already vulnerable to interference from space weather and human activity, and no critical service relies on it alone—and, in the authors’ view, the problem of synchronising satellites can perfectly well be solved within that industry, without shifting the burden onto everyone else.
The argument did not stay behind closed doors. In 2022, at the 27th session of the Conference, Russia voted against abolition—its GLONASS system uses leap seconds, and the transition would have required replacing satellites and ground stations; Moscow proposed postponing the decision until 2040 or later. Canada, the United States, and France, by contrast, pressed for haste and wanted the matter settled before 2035. The compromise everyone accepted was elegant and bureaucratically precise: abolition was decided in principle, while the exact date and figure were deferred to the next session.
Now for what will happen in Versailles. Draft Resolution C raises the threshold for the discrepancy |UT1 − UTC| to 3600 seconds, that is, a full hour. According to the committee’s calculations, this ensures continuity of the scale for “several centuries”—elsewhere in the document, they already speak of “approximately a millennium.” The date proposed by the committee and approved by the CIPM is 20 May 2027.
Why an hour? The working group considered three options: a minute gave roughly a century of continuity, no limit meant “possibly never,” and an hour won as the compromise. Systems that transmit the value UT1 − UTC to users need a known maximum in order to allocate the digits correctly in the service field—“never” does not work for them, and the document says this outright. Radio stations that still broadcast the correction have been given a transition period until 2035, or until 2040 if the technology does not allow it. Though according to a 2025 survey, only five countries still have such stations: Argentina, Canada, Russia, the United Kingdom, and the United States.
And there is one final irony that closes the story. The resolution is being pushed forward precisely to avoid encountering a negative leap second: according to 2025 models, its probability in the next decade is around 30%, while the probability of a positive second is around 50%. In other words, they are abolishing a practice that would be needed more often than usual in the years ahead. Humanity has more or less learned to survive a positive second; a negative one, it has not, and the industry considers the likelihood of error in this untested scenario too expensive.
The answer to the question in the hook is therefore dry: continuous UTC is a choice of the devil you know, not a victory of precision over nature. The planet will keep rotating as it pleases, and the gap between sky and clocks will not go away—it will simply stop being someone’s daily problem. Over a century, the clocks will run ahead by several minutes. The only question is who will notice: from 2027 onward, almost nobody.