Leap Seconds Explained: Why the Clock Sometimes Says 23:59:60

Atomic clocks are steadier than the planet they sit on. The second was fixed by definition in 1967, but the day it was meant to divide keeps changing length, so the two slowly pull apart. A leap second is the patch: an extra second inserted into Coordinated Universal Time to drag it back into step with the turning Earth, and the cause of a surprising amount of trouble for such a short interval.

This article explains why Earth’s rotation makes the correction necessary, who decides when one is applied, the damage it does in computing, and why leap seconds are being abandoned before 2035. The scale it adjusts can be read on the current UTC clock.

What Is a Leap Second?

It is an extra second added to UTC to keep it close to the time told by Earth’s rotation. When one is inserted, the last minute of the chosen day runs to sixty-one seconds and clocks read 23:59:60 before rolling over to midnight.

UTC is built from atomic time but is required to stay within 0.9 seconds of UT1, the timescale defined by Earth’s actual rotation angle. Whenever the difference threatens to exceed that limit, a leap second is inserted to close the gap. Twenty-seven leap seconds have been added since the system began in 1972, every one of them positive, and the most recent arrived at the end of 31 December 2016. That is why International Atomic Time now runs 37 seconds ahead of UTC: ten seconds of initial offset, plus the twenty-seven corrections applied since. Both scales tick at the same rate; only the count of whole seconds separates them.

Why Earth’s Rotation Forces a Leap Second

The planet is a poor clock. Tidal friction from the Moon has been slowing its spin for as long as there has been a Moon, lengthening the mean solar day by roughly two milliseconds per century, and the SI second was defined to match a day of the nineteenth century rather than today’s.

A modern mean solar day is therefore a shade longer than 86,400 SI seconds. That shortfall is invisible on any single day and adds up relentlessly: a millisecond or two lost daily becomes most of a second within a couple of years. On top of the steady tidal slowing sit irregular changes, some of them large enough to swamp it. The exchange of angular momentum between the atmosphere and the solid Earth shifts the rotation rate seasonally, and movements in the molten core, melting ice sheets and the rebound of land once buried under glaciers all leave their mark. Since around 2020 the planet has been rotating slightly faster than it had for decades, which is why no leap second has been required since 2016 and why a negative one, removing a second rather than adding it, has moved from a theoretical curiosity to a real possibility.

Who Decides When One Happens?

The International Earth Rotation and Reference Systems Service, the IERS, makes the call. It measures the planet’s rotation, compares UT1 with UTC, and announces any leap second in a notice known as Bulletin C, normally about six months in advance.

The preferred dates are the ends of June and December, with the ends of March and September available if needed. Notice is deliberately generous because a great deal of equipment has to be told, from broadcast networks to satellite ground stations. The decision cannot be made much earlier, however: the Earth’s rotation is not predictable far ahead, so nobody knows in 2026 whether a correction will be wanted in 2029. That combination, mandatory and yet unschedulable, is the root of the engineering objection to leap seconds.

The Trouble Leap Seconds Cause in Computing

Computers are built on the assumption that every day contains exactly 86,400 seconds. Unix time encodes that assumption directly, which leaves no way to represent 23:59:60, so the extra second is usually handled by repeating a value or freezing the clock.

The consequences are not theoretical. The leap second of June 2012 triggered a kernel fault that sent processes on many Linux servers into a spin, taking down well-known websites and an airline check-in system for hours. Repeated timestamps break assumptions that databases and log analysis rely on, ordering can invert, and any code that subtracts one time from another can produce a negative interval. Several responses are now in common use:

• Repeat the second: the strictly correct behaviour, where the same Unix timestamp occurs twice in succession. Simple, standards-compliant and hazardous for anything that assumes timestamps are unique.
• Freeze or step: hold the clock still for a second, or step it backwards. Backwards motion is the most damaging option of all for software.
• Smear the second: spread the correction over several hours by running clocks fractionally slow, so no client ever sees a discontinuity. Large cloud providers do this.
• Ignore it and resynchronise: let the machine fall a second out and allow the Network Time Protocol to pull it back afterwards, which is what most unattended equipment effectively does.

Smearing solves the problem locally and creates a new one globally, because a smeared clock and a strict clock disagree by up to a second during the smear window. The protocol layer’s role in all this is covered in what NTP is and how it works.

The Decision to Phase Them Out

In 2022 the General Conference on Weights and Measures resolved to stop inserting leap seconds by or before 2035. The tolerance between UTC and Earth rotation time will be widened to a value still to be agreed, and until then the existing rules continue unchanged.

The argument that carried the day was practical. Leap seconds are rare, unpredictable, expensive to implement correctly and, when implemented incorrectly, capable of taking down infrastructure, while almost nobody outside astronomy and satellite navigation needs civil time tied to the sun to better than a second. Once the practice stops, UTC will drift away from solar noon at the rate of about a minute per century, a discrepancy no ordinary user will ever notice and which a future generation can address with a larger, scheduled adjustment. Astronomers and telescope operators, who genuinely need UT1, will continue to obtain it as a published correction rather than having it baked into civil time. The offsets that result are set out in TAI vs UTC vs GPS time, and the laboratories that will implement the change in national time standards.

Conclusion

A leap second exists because an atomic definition of the second and a slowing, wobbling planet cannot stay aligned on their own. The IERS announces each insertion around six months ahead, the extra second appears as 23:59:60 at the end of June or December, and computers have never handled it gracefully. Twenty-seven leap seconds have been added since 1972, the last in 2016, and the practice ends by 2035. To see the scale these corrections apply to, open the live UTC clock, compare it with atomic time on the TAI comparison page, or start from the atomicclock.now homepage.