In July 2026, an engineer posted a question to Hacker News: “It’s been 10 years since we had the last Leap secondAn occasional one-second adjustment to Coordinated Universal Time (UTC) that keeps it within 0.9 seconds of time based on Earth's rotation.Read more on Wikipedia ↗ and it looks like we will get the first negative one soonish. Are systems ready for that?” The rounding is off — the last leap second arrived at the end of December 2016, nine and a half years earlier — but the question underneath is dead serious, and nobody on Earth can answer it from experience. In fifty-four years of leap seconds, a negative one has never happened.
A minute with 59 seconds
The familiar kind of leap second adds. A clock that normally rolls from 23:59:59 to midnight instead holds 23:59:60 for a beat, and the day is one second longer. Every one of the 27 leap seconds inserted since 1972 has worked this way, which is why Coordinated Universal Time (UTC)The world's reference time scale, kept by atomic clocks and held within a second of Earth's rotation; time zones are stated as offsets from it.Read more on Wikipedia ↗ (UTC) now runs 37 seconds behind atomic time.
The negative kind subtracts. The standard that defines the system — the International Telecommunication Union’s (ITU) time signals and frequency standards emissions series (TF) Recommendation TF.460-6 — spells out the procedure plainly. In the case of a negative leap-second, 23h 59m 58s will be followed one second later by 0h 0m 0s of the first day of the following month. The final minute of the day has 59 seconds. 23:59:59 never happens — not repeated, not shown twice, just skipped.
Everything else follows the old rules. It would land at the end of a UTC month, December or June preferred. The International Earth Rotation and Reference Systems Service (IERS)The scientific service that tracks Earth's rotation and announces, in its Bulletin C, when a leap second is needed.Read more on Wikipedia ↗ (IERS), the service that tracks Earth’s rotation, would announce it at least eight weeks ahead — the announcement channel is the same six-monthly bulletin that routinely says no.
Why the planet nearly demanded one
Leap seconds exist because the Earth’s spin drifts against atomic time. When the difference builds toward 0.9 seconds, a leap second resets it. The spin has been slowing for centuries, so the corrections have always been insertions. Then, around 2020, the drift went the other way.
Days began running short. Every year from 2021 through 2024 broke the pre-2020 record, and the shortest of all — 1.66 milliseconds under 86,400 — came on 5 July 2024. A geophysicist at the Scripps Institution of Oceanography, Duncan Agnew, put the implication in Nature in March 2024: extrapolating the trends, UTC as now defined will require a negative discontinuity by 2029. That is a projection, not a schedule — and 2029 is the paper’s word, conditional on the trends holding.
They may not be holding. 2025’s shortest days came in around 1.3 milliseconds under, short of the record and still marked “to be confirmed” by the data compiler; one rotation specialist, Leonid Zotov, read the numbers as the Earth having started to decelerate again. Live data agrees: the broadcast correction has sat at 0.0 since April 2026, and the rapid solution put the Earth-atomic difference at just over a millisecond in early September — nowhere near the trigger. Expert opinion is genuinely split. Agnew’s extrapolation says 2029; the projections of Demetrios Matsakis, relayed by the Cambridge time expert Tony Finch, suggest we might not have a negative leap second at all.
The climate fact everyone repeats backwards
One more twist, because the pop coverage of this story keeps getting it upside down. Melting ice in Greenland and Antarctica, measured by satellite gravity, has decreased the Angular velocityHow fast something rotates, measured as the angle it turns through per unit of time.Read more on Wikipedia ↗ of the Earth. Melting polar ice slows the spin — mass shifting toward the equator — rather than speeding it up. Agnew’s own arithmetic: without the accelerated melt, the problem would have arrived about three years earlier. Global warming did not cause the negative leap second. It postponed it.
What did speed the planet up, in Agnew’s account, is deeper: since 1972 the angular velocity of the Earth’s liquid core has been decreasing at a rate that has steadily increased the angular velocity of the rest of the planet.
The race
Here is the strange part. The geophysics says maybe. The paperwork is moving faster.
In October 2026 — the 13th through the 15th, in Versailles — the general conference on weights and measures votes on a draft resolution that would make UTC continuous. No leap seconds of either sign, ever again, effective 20 May 2027, with the allowed Earth-atomic difference raised to a full hour. The preamble does not mince words about why. A 2025 workshop of time and Earth-rotation experts put the probability of a negative leap second at 30 percent by 2035. The preparation cost for industry is estimated as similar to the millennium bug. International bodies and industry representatives formally asked for the date to be brought forward specifically to avoid the negative leap second. And the resolution’s own language calls the infrastructure it would stress largely unprepared.
Note what this is not. The 2022 decision that everyone remembers as “abolishing leap seconds” decided only that the tolerance would be increased in, or before, 2035 — it abolished nothing. The October vote is the actual abolition, and it exists because of a second that has never been skipped. Nature’s news desk framed it flatly: fear of a possible negative leap second is accelerating plans to abolish the adjustment. Agnew saw it coming — his paper warned the problem might require changes in UTC earlier than planned.
If the vote passes, the negative leap second is cancelled before any kernel has to skip a beat. The first leap-second crisis would be averted by abolishing the mechanism rather than fixing the software.
What a deleted second does to software
The engineering record is a strange mix of specified and never exercised.
POSIX (the Portable Operating System Interface, the standard behind Unix time) is the bluntest case. Leap seconds are ignored — not applied — and every day is exactly 86,400 seconds by definition. The standard’s own rationale admits that broken-down POSIX time is not necessarily UTC, despite its appearance. A day with a second removed is simply outside what the seconds-since-the-Epoch mapping represents.
The protocols know the concept. The Network Time Protocol (NTP) has carried a code for it since 2010: leap indicator 2 means the last minute of the day has 59 seconds. Linux has a kernel interface flag for deleting a leap second at the last second of the UTC day, and chrony documents the path its default mode would take. When deleting a leap second, it steps forward by one second when the clock gets to 23:59:59 UTC. Google’s smear design extends to the negative case too — on paper, speeding clocks up over the day instead of slowing them.
Specified, in other words, at every layer. Exercised against reality, at none. The standards bodies’ own words: the insertion has never been foreseen or tested.
The direction matters, and most descriptions get it backwards. A negative leap second makes time jump forward, not backwards — 23:59:58 to midnight with nothing between. It is the familiar positive leap second that makes naive wall-clock time appear to run in reverse. That inversion is not a technicality; it is where the damage has actually been. When Cloudflare’s Domain Name System (DNS) failed on New Year’s Day 2017, the root cause was code built on the belief that time cannot go backwards. A positive leap second, smoothed by upstream servers, produced a negative time difference and crashed a random-number call. A negative leap second stresses different assumptions, and one Global Positioning System (GPS) detail is worth naming: GPS time is continuous, so a deleted second would move the GPS-UTC offset from 18 seconds to 17 — harmless over the air, risky wherever 18 is hardcoded.
Ready or not
So are systems ready? The honest answer is that informed people disagree, loudly.
One side: the writer qntm’s much-shared 2024 essay — “No one, or almost no one, is ready for a negative leap second” — and the institutional voices. The BIPM’s Time Department director, Patrizia Tavella, has called the problems without precedent. Agnew: many systems can accept an additional second, but few if any allow for removing one. Meta’s engineering blog warned of a devastating effect on software relying on timers or schedulers.
The other side, from the practitioners who would carry it: the negative case is arguably the easier one, because time only ever moves forward. On the Internet Assigned Numbers Authority (IANA) time-zone mailing list, Brooks Harris put it that way. On the Hacker News thread, one commenter’s summary was that badly coded things would report an operation took 1100 milliseconds instead of 100 — not great, but not minus 900.
Both sides are describing the same untested event. The difference is which failure they fear more: a forward jump no one has rehearsed, or abolishing the correction that keeps clock time tied to the sky.
Which is the real stakes of the October vote. The planet spent four years spinning faster than the clock, and the response of the world’s timekeepers was not to fix the software — it was to change what the clock is for. If you want how we got here, the leap second’s whole story is its own post; if you are timing something yourself, a duration needs no meridian and no bulletin — start it, and count.
Sources
- IERS Bulletin C 72 (July 2026) and Bulletin A Vol. XXXIX No. 036 — the no-December-leap announcement, UTC−TAI (International Atomic Time) = −37 s, DUT1 0.0 s since April 2026.
- Recommendation ITU-R TF.460-6 — the negative-leap procedure (23h 59m 58s → 0h 0m 0s), scheduling and the eight-week notice rule.
- BIPM: Resolution 4 of the 27th CGPM (2022) and the Draft Resolutions of the 28th CGPM (October 2026) — “never been foreseen or tested”, the 30-percent-by-2035 estimate, Y2K-scale cost, continuous UTC from 20 May 2027.
- Duncan Agnew, “A global timekeeping problem postponed by global warming”, Nature 628 (2024) — the 2029 projection, the ice-melt delay, the core mechanism, and the prediction the vote now answers.
- timeanddate.com leap-second pages (data compiler citing IERS and the US Naval Observatory) — 27 positive insertions, the record-day series, the 2025 “to be confirmed” figures, Zotov quote.
- Request for Comments (RFC) 5905 (NTPv4), POSIX rationale A.4.14 and §4.16, chrony.conf(5) 4.6.1, adjtimex(2) man page — the specified-but-unexercised stack.
- Google, “Leap Smear” (vendor page, own practice only) — the negative-case smear as design intent.
- Hacker News threads via Algolia (snapshot 7 September 2026) and the IANA tz mailing list — the Ask Hacker News question, practitioner consensus, Harris and thomashabets2 quotes.
- qntm, “All I want for Christmas is a negative leap second” (2024); Tony Finch, “An update on leap seconds” (2022) — the contested readiness verdict and the Matsakis counter-projection.
- Cloudflare post-mortem (2017) and Meta Engineering (2022, operator pages on their own incidents); NPR (2024) quoting Tavella and Agnew; Nature news (31 August 2026) on the October vote.