The leap second is retired: UTC goes continuous on 20 May 2027, with a one-hour cap
The 28th CGPM passed Resolution 5 on 15 October in Versailles: continuous UTC from 20 May 2027, with UT1-UTC allowed to drift up to one hour.
The General Conference on Weights and Measures (CGPM) adopted Resolution 5 of its 28th meeting on 15 October 2026 in Versailles: from 20 May 2027, Coordinated Universal Time stops inserting leap seconds to stay aligned with Earth’s rotation (UT1), and atomic time is allowed to drift away from solar time until the gap reaches a full hour. The vote belongs to the CGPM, the supreme body of member-state representatives under the Metre Convention; the International Bureau of Weights and Measures (BIPM) is its secretariat and executive arm, and the lab that actually produces UTC.
Key points
- The date: continuous UTC takes effect on 20 May 2027 — World Metrology Day. Resolution 5 sets the maximum |UT1-UTC| at 3,600 seconds (one hour).
- Why now: the resolution cites a joint CCTF/IERS workshop in March 2025 estimating the probability of a negative leap second at 30% by 2035, and calls the preparation cost comparable to millennium-bug remediation.
- The schedule moved up eight years: Resolution 4 of the 27th CGPM in 2022 only asked for a larger cap “in, or before, 2035.” This resolution sets the switchover for 2027.
What the resolution actually says
Resolution 5 is titled “On the technical actions needed to ensure the continuity of UTC,” and its body runs through recalling, recognizing, further recognizing, noting, welcoming, decides and requests clauses. Only two decides lines bind: “continuous UTC will become effective on 20 May 2027,” and “the maximum value for the difference |UT1-UTC| will be 3 600 seconds (1 hour), ensuring the long term continuity for UTC for several centuries.” CCTF president Noël Dimarcq put a longer horizon on it in his closing statement: the framework “is expected to maintain the continuity of UTC for several millennia.”
The requests clause hands the BIPM four jobs: keep collaborating with the ITU, IAU, IUGG, IERS and others to inform users; keep monitoring UT1-UTC and its modelled evolution; report regularly to the CIPM; and work with regional metrology organizations, national metrology institutes and users on understanding and delivering continuous UTC. Nothing in the text tells anyone to change code by a given date. It sets the time rule and a reporting mechanism, and leaves the engineering to everyone else.
The resolution also argues its own case for a one-hour tolerance: in many countries local apparent solar time already differs from civil time by more than an hour because of time-zone choices, and an hour’s jump happens every year anyway when daylight saving time starts. Civil time and the Sun were never locked to the minute.
Twenty-one years of argument
Leap seconds arrived in 1972, when UTC was already 10 seconds behind TAI from earlier methods. Twenty-seven were inserted in total, all of them positive; the last was 31 December 2016, so UTC has sat 37 seconds behind TAI since January 2017. The campaign to delete them is older than most people assume: on 5 July 2005 the United States proposed to ITU-R that leap seconds be eliminated before 2008, with leap hours as the technical replacement, partly because several ITU-R member states legally require civil time to track the Sun. That November’s discussion was postponed, the 2012 vote was deferred, and WRC-15 decided to keep studying.
On 18 November 2022 the 27th CGPM passed Resolution 4, raising the |UT1-UTC| cap “in, or before, 2035” to keep UTC continuous for at least a century. The 2035 date was a concession to Russia, which asked for 2040 because GLONASS follows leap seconds while GPS does not. WRC-23, which closed on 15 December 2023 in Dubai, formally recognized that direction through Resolution 655, while leaving a transition path: equipment that cannot be replaced sooner may keep receiving the larger UT1-UTC difference over radiocommunication systems until 2035, and no later than 2040. Resolution 5 quotes it, which means old and new practice will overlap for years after 2027.
What forced the issue
Earth’s rotation. The planet spun faster in 2020, producing the 28 shortest days recorded since 1960, and 29 June 2022 was the shortest day on record, 1.59 milliseconds under 24 hours. A 2024 Nature paper by Duncan Agnew of the Scripps Institution of Oceanography projects that meltwater migrating from the ice caps toward the equator will slow the rotation again. The uncertainty itself is the risk: if the trend runs the other way, the first thing humanity needs is a negative leap second, one that never been tried.
The resolution’s language on that is blunt: a negative leap second “has not previously been applied, and it is considered to pose a high risk of causing anomalies and disruption to critical infrastructures that are largely unprepared,” with preparation costs “estimated to be similar to their preparations for the millennium bug.” The same clause names IoT, dispersed multi-sensor systems, smart grids and virtual services as the reasons the risk keeps growing. Another clause is purely engineering: global providers each smear the leap second differently, so their clocks disagree with each other and the time they disseminate no longer conforms to the SI second.
There is a political-economy clause too. If UTC stays discontinuous, the resolution warns, some user organizations will switch their reference time scale to GNSS time instead, “seriously compromising the concept of international standardization in metrology.” The Verge lists Meta, Google, Microsoft and Amazon as companies that “have all supported efforts to scrap the system.” Only Google and Amazon show up in Wikipedia’s account of smearing — Google stretching the second over 24 hours, Amazon following a similar pattern from the June 2015 insertion and later shipping an EC2 NTP service that smears — so the supporter list and the smearing list are not the same four names.
Leap seconds really did break things
Wikipedia’s entry records the 30 June 2012 insertion taking down Reddit (Apache Cassandra), Mozilla (Hadoop), Qantas and a range of Linux sites. Reddit is the most-cited name on that list; we covered it again on 30 September when it shut down RSS and killed its public API. The 30 June 2015 one went wider: Twitter, Instagram, Pinterest, Netflix, Amazon and Apple’s Beats 1 all saw interruptions; Linux leap-second bugs hit the Amadeus Altéa reservation system used by Qantas and Virgin Australia; Cloudflare’s DNS resolver computed a negative number by subtracting two Go time.Now() timestamps, which Go 1.9 fixed by switching to a monotonic clock source. The most expensive incident was in finance: the Intercontinental Exchange, parent of seven clearing houses and eleven exchanges including the NYSE, closed its markets for 61 minutes across the 2015 leap second.
The Verge names Foursquare, Reddit, LinkedIn and Yelp. The two lists don’t match exactly, and that is not a contradiction — different reporting quoted different incident logs. What is solid is the range: social networks, travel, finance, aviation and cloud providers, between 2012 and 2015.
The seven months in between
Resolution 5 lists no migration checklist, only “inform users and support the application of this decision.” Read the recognizing, further recognizing and noting clauses together, though, and the exposed surface is clear:
| Who is affected | What the resolution flags |
|---|---|
| GNSS-timed systems | ITU-R Working Party 7A is studying code updates to broadcast the larger UT1-UTC difference, with a transition period to 2035/2040 |
| Transport, energy, telecoms, space | Named in the text as critical infrastructure that needs UTC continuous with no future adjustment |
| Software that needs UT1 | Must now read UT1-UTC from GNSS navigation messages, IERS or national metrology institute services |
| Time-sensitive protocols | IEEE 1588 and others have formally asked the BIPM to bring continuous UTC forward |
The rhythm of the whole affair: direction set in November 2022, endorsed at WRC-23 in December 2023, a fixed switchover date in October 2026, and the flip on 20 May 2027. Between the vote and the switch there are seven months for ITU-R to update codes and for industries to adjust software — a fraction of the runway the 2035 plan allowed.
Our take
The direction is right for anyone running distributed systems, but two misreadings are worth heading off. The first is that time becomes a non-problem. It does not: what disappears is the irregular second. A one-hour cap pushes the correction interval into the millennium range — The Verge reports the full hour is “not expected to happen for around 3,000 years,” while Wikipedia records a proposal to let the gap grow to a full minute first, which “would take 50 to 100 years,” and then stretch the last minute of the day into two. What form a correction takes is left open by the resolution. The second is that solar time stops mattering. Astronomy, navigation and timing services still need UT1; they will just fetch it from GNSS messages or a dedicated NTP service instead of reading an IERS bulletin.
For cloud and infrastructure teams the practical move is to audit the time path: how NTP clients and servers handle leap seconds today, where code quietly assumes UTC and UT1 stay within a second of each other, and where timestamps are used for ordering. Cloudflare’s bug is the concrete lesson — subtract timestamps against a monotonic clock. None of that is urgent, but software written after May 2027 should not carry the old assumption at all.
Three things stay open. What a one-hour correction actually looks like — the resolution only asks the BIPM to keep monitoring and report to the CIPM. How fast ITU-R gets its new broadcast codes out. And which way Earth’s rotation goes between now and 2027, which decides whether the world faces a first, unrehearsed negative leap second before continuous UTC even starts.