Leap seconds end on 20 May 2027, and the Earth has already turned around

Since 1972, the world's clocks have been kept in step with the Earth's rotation by inserting an extra second now and then. Twenty-seven times, a minute has run to 61 seconds. A draft resolution proposes that on 20 May 2027 this stops. The decision itself is one line — "continuous UTC will become effective on 20 May 2027" — but the timing is what's interesting: the committee chose to stop predicting the Earth's rotation just as it stopped doing what the predictions said.

What is actually being decided, and by whom

The text is Draft Resolution C of the 28th General Conference on Weights and Measures, "On the technical actions needed to ensure the continuity of UTC". Two things sit under its decides that heading:

• Continuous UTC takes effect 20 May 2027.
• The tolerance on the gap between clock time and Earth time — |UT1−UTC| — rises to 3,600 seconds, one hour, described as "ensuring the long-term continuity for UTC for several centuries".

⚠️ It is a draft, and the vote has not happened. The conference sits from 13 to 15 October 2026 in Versailles. The date has been fixed in the text since version 2 on 30 January 2026 and no later version has touched Draft Resolution C, so it is stable rather than provisional — but anyone writing "leap seconds have been abolished" today is a few weeks ahead of the record.

An hour of drift sounds enormous until you notice the resolution's own justification: in many countries the gap between local clock time and actual solar time is already more than an hour by choice of time zone, and daylight saving moves it by an hour twice a year on purpose. Civil time has never tracked the sun closely. It only ever tracked it to within a second because a committee kept forcing it to.

The number everyone quotes is a recital, not a decision

The figure in circulation is a 30% probability of a negative leap second by 2035 — a minute of 59 seconds rather than 61, which has never been done. It comes from a March 2025 workshop of Earth-rotation experts convened by the Consultative Committee for Time and Frequency and the IERS.

⚠️ In the resolution that estimate sits under noting that, not under decides that. It is context the committee recorded, not a finding it adopted. That distinction gets lost every time the number is repeated. It matters because the 30% figure is a reason the committee gave for acting, not a prediction the committee itself made.

Why a negative leap second is the frightening case is set out plainly in the same document. It has never been applied; it is judged to pose "a high risk of causing anomalies and disruption to critical infrastructures that are largely unprepared"; and preparing for it would cost the time-synchronisation industry an amount "similar to their preparations for the millennium bug". A quieter risk is also named: further delay could push organisations to abandon UTC for a GNSS time scale, "seriously compromising the concept of international standardization in metrology".

We ran the projection ourselves, and it is obsolete

The obvious thing to do with a decision like this is check the underlying trend. The Earth's rotation is published: the IERS Earth Orientation Parameters series gives UT1−UTC daily, back to 1962.

Take the change on 1 January of each year:

The recent mean is about +31.3 ms a year. UT1−UTC stood at +0.0741 s on 1 January 2026, and a negative leap second is triggered when it climbs toward +0.9 s. Divide, and the answer is roughly 26 years: mid-2052.

That number is arithmetically correct and physically worthless, and we are printing it only to explain why.

Two things the projection cannot see

The first is the dataset's own edge. The series ends on 5 January 2026. The last interval above is not a choice; it is the last one the file can supply. Nothing about the file announces this, and an extrapolation run against it inherits a picture eight months out of date without saying so.

The second is what happened in those eight months. On 27 August 2026, UT1−UTC stood at +0.006 s — down from +0.074 s in January. A fall of about 68 milliseconds in a series that had been rising for five years.

The tempting explanation is the seasonal term. The Earth does speed up and slow down through the year, and any single reading contains that wobble. Apply the standard seasonal correction and about 14 milliseconds of the fall is seasonal. The other 82 milliseconds are real — a deseasonalised rate of roughly −126 ms a year, against the +31 ms a year the projection was built on.

⚠️ This is not our interpretation against the IERS's. In its own working prediction, Bulletin A of 27 August 2026, the linear coefficient is −0.00022 seconds per day. That works out to −80.4 ms a year — a negative trend. Its own daily table has UT1−UTC crossing zero around 8 September 2026 and continuing down. The observations and the operational model agree with each other, and both disagree with the straight line.

Which is the actual argument for the change

A trend that had run one way for five years reversed sign within eight months of the last data point a careful person could have used. Not a small revision — a change of direction, on the quantity the entire leap-second mechanism keys on.

That is the case for removing the trigger rather than forecasting it. The committee is not claiming to know what the Earth will do. It is declining to make the world's timekeeping depend on knowing.

Read that way, the hour-wide tolerance is not carelessness about accuracy. It is a decision to stop issuing short-notice corrections against a signal that changes faster than the corrections can be prepared for.

What actually changes for you

Almost nothing, and that is the point. Unix time keeps counting, timestamps keep parsing, and the difference between civil time and solar time grows at a rate you will not notice in a lifetime — the resolution's own framing is that an hour of tolerance lasts "several centuries".

What goes away is the thing that has actually broken systems: the irregular, short-notice insertion of a second that some software handles by repeating a timestamp, some by smearing it across a day, and some by crashing. If your system has ever had a leap-second incident, the fix is the removal of the event, not a better handler for it.

⚠️ One thing does not go away by 2027. ITU-R Resolution 655 allows the growing UT1−UTC difference to keep being disseminated over radio "until 2035, but no later than 2040, in cases where existing equipment cannot be replaced earlier". Anything reading time off a radio signal has a longer horizon than the headline date suggests.

If you work with timestamps across zones, our Unix timestamp converter and timezone converter both treat UTC as the reference the way software does, which is exactly the convention this decision preserves. For intervals in days rather than seconds, the date difference calculator is unaffected by any of this — leap seconds have never been visible at that resolution, which is another way of saying how small the quantity under discussion is.

Where this comes from, and what will date it

The date, the tolerance, the 30% estimate, the millennium-bug comparison, the GNSS warning and the ITU-R deadline were all read from the CGPM draft resolutions document itself — version 5, 13 July 2026 — not from coverage of it. The rotation figures were computed from the IERS Earth Orientation Parameters series, and the prediction coefficient read from IERS Bulletin A of 27 August 2026. The arithmetic above is reproducible from those two published files.

Two limits worth stating. We did not attempt to say when a negative leap second will be needed — the whole point of the section above is that we showed our own attempt to be unsound, and a second attempt would inherit the same problem. And the resolution is a draft: the vote is in October 2026, so the correct present-tense statement is that this is what is proposed and stable, not what has been enacted.

This guide dates on 15 October 2026, when the conference either adopts the text or does not.