At 23:59:59 UTC on a chosen day, the next minute stretches to 61 seconds. The display reads 23:59:60 before rolling to 00:00:00. That extra pulse is a leap second, a correction inserted into Coordinated Universal Time to stop atomic clocks drifting away from the sun. Without it, the gap between precise caesium ticks and Earth's lazy rotation would widen by roughly one second every 18 months.

How a leap second works

A normal minute has 60 seconds. A leap second adds one more, almost always at the end of June or December. The sequence runs:

23:59:59
23:59:60
00:00:00

The insertion lands just before midnight UTC. Clocks that follow UTC (smartphones, computers, network time servers) display that extra tick. Clocks set to local time show it at whatever hour corresponds to UTC midnight in that zone.

Since 1972, the International Earth Rotation Service (IERS) has added 27 leap seconds. The most recent was on 31 December 2016. As of 2026, none have been added since.

Why does UTC need leap seconds?

Two clocks are running.

Atomic time (TAI) is kept by caesium clocks. It never wavers. The SI second is defined by a fixed number of oscillations of a caesium-133 atom. Atomic clocks do not slow down.

Astronomical time (UT1) is read from Earth's rotation. One spin equals one day, but the spin is not steady. Tidal friction from the Moon brakes Earth by about 1.7 milliseconds per century. Small, but cumulative. Without correction, atomic noon would eventually arrive while the sun still sits low in the sky.

Leap seconds are the patch that keeps the two clocks aligned.

How does the IERS decide on a leap second?

The IERS tracks Earth's rotation with radio telescopes and laser ranging to satellites. When the gap between atomic time and astronomical time nears 0.9 seconds, the IERS issues a bulletin calling for a leap second. The insertion is scheduled for 30 June or 31 December.

The decision is human, not automated. The bulletin typically arrives about six months ahead.

What happens during a leap second

Most people notice nothing. Their device pulls time from a network server that absorbs the extra tick.

Some systems break. In 2012, a leap second triggered outages at Reddit, Mozilla, LinkedIn, and assorted Linux servers. The culprit was a timing bug: software that assumed every minute held exactly 60 seconds encountered a state it could not parse. Servers froze, CPUs spiked, services dropped.

Since then, large tech operators have adopted a "smear" strategy. Instead of inserting a whole second at once, they stretch the final 1,000 milliseconds of the day across 2,000 milliseconds. The clock runs slightly slow for a short window. Nothing ever sees a 61st tick. Google, Amazon, and Meta all use this method.

The plan to abolish leap seconds

Leap seconds are a maintenance burden. They are unpredictable. They occur at most twice a year, but the exact date is unknown far in advance. Every system that touches time must account for them, and many get it wrong.

In 2022, the International Bureau of Weights and Measures resolved to stop adding leap seconds by 2035. The World Radiocommunication Conference confirmed the plan. After 2035, UTC will cease inserting new leap seconds. The drift between atomic time and astronomical time will be allowed to grow, managed separately for applications that require solar alignment.

The 27 leap seconds already added stay baked into UTC. No more will follow. The clock will tick uniformly from that point forward. For most people, nothing changes. For astronomers and satellite operators, the accumulating offset will eventually be handled by a note in the time standard rather than by halting the clock.

What leap seconds mean for your systems

If you maintain software that handles time, support leap seconds until at least 2035. After that, you can stop. The IANA time zone database ships leap second data in a file called leapseconds. Tools like Python's zoneinfo and Java's tzupdater read it. If your application cannot handle a 61st tick, test it with the smear method or switch to a library that can.

If you are a regular user, you will never see a leap second on a smartphone or laptop. The operating system absorbs it. Do not worry about it.

When leap seconds break your code

Software developers who build scheduling applications, calendar tools, or distributed databases should test their code against a leap second before 2035. The default advice is "let the OS handle it." That works for most apps. It fails for systems that count elapsed time in seconds or log timestamps with sub-second precision. If your system stores time as a count of seconds since an epoch (Unix time, for example), a leap second can cause that number to repeat or skip. Test for it.

After 2035, the problem disappears. That is the point.