Leap second abolition ahead: NTP, GPS and trading systems face a fixed UTC offset

Two articles published on dev.to report that the International Bureau of Weights and Measures (BIPM) has moved to abolish leap seconds, ending the practice of inserting one-second adjustments into Coordinated Universal Time. The English-language piece dates the announcement to May 12, 2026 and says leap seconds will be abolished after 2027. A Spanish companion article by the same author gives 2035 as the year the change takes effect. The sources disagree on the timeline, but they agree on what follows: every time-synchronisation stack that depends on dynamic leap-second tables has to be audited, patched and re-verified.

What actually changes

Leap seconds exist to keep UTC aligned with the irregular rotation of the Earth. According to the dev.to articles, the offset between UTC and the continuous atomic timescale TAI currently stands at 37 seconds, with a leap second added roughly every 18 months. After the abolition, that offset is frozen at 37 seconds permanently. The NTP leap indicator will always read zero, leap-smear windows on servers that spread the adjustment over 24 hours disappear, and a timestamp of 23:59:60 will never be issued again.

The Spanish article spells out the failure mode for unprepared systems: software that reads the IERS leap-second list keeps working only while that list keeps being updated. Once updates stop, every omitted leap second leaves dependent clocks about one second out of step, producing incorrect timestamps, broken event correlation and, in some cases, failed digital-signature validation.

A disputed cut-off date

The two pieces, published the same day by the same author, do not agree on when the change lands. One says BIPM announced on May 12, 2026 that leap seconds end after 2027; the other says the disappearance is definitive from 2035. The formal BIPM decision text should be treated as the authority on the date, but the migration work is identical either way, and planning against the earlier date is the safer option.

Who is affected

Network time is the first concern: NTP and PTP servers that implement leap smearing must stop advertising a non-zero leap indicator and lock in the fixed offset. Both articles single out leap-file dependencies in ntpd and chrony configurations as audit targets.

GPS devices form the second group. GPS time already runs ahead of UTC by 18 seconds as of 2024, per the Spanish article, and that delta becomes constant once leap seconds stop. Receivers that convert GPS time to UTC using leap-second tables need firmware updates to use the fixed constant instead.

Financial markets face regulatory exposure. Rules such as MiFID II and SEC requirements demand high-precision timestamps, and according to the articles a one-second skew can produce out-of-order trades, failed reconciliation and penalties, while a static conversion removes a recurring source of reconciliation errors. Distributed ledgers face fork risk from a sudden jump in consensus timestamps, and embedded or IoT devices with hard-coded leap-second tables in their RTC chips may freeze or roll over to invalid dates once the expected entry never arrives.

A migration checklist distilled from both articles

Inventory every time source: NTP, PTP, GPS receivers, RTC chips and software clocks. Search configurations for leap-second table references such as leap-seconds.list. Replace lookup tables with the constant 37-second offset in all conversion code. Remove any scheduled jobs that adjust clocks around leap-second events. Patch firmware where vendors support it, and where they do not, isolate devices behind a gateway that strips the leap flag. Regression-test by generating timestamps across a simulated midnight transition and verifying they stay monotonic. Finally, add drift monitoring, with the Spanish article suggesting alerts when offset exceeds 10 milliseconds, and document a post-cutover runbook.

Why it matters

Timestamps are the ordering primitive underneath databases, logs, audit trails and signatures; when clocks disagree by a second, event correlation and ordering guarantees break quietly rather than loudly. Abolishing leap seconds removes a recurring operational hazard, and the English article attributes the decision to the mounting cost of maintaining them in software updates, testing and operational risk. But the benefit only materialises for teams that complete the migration. The real exposure sits in the long tail of unpatchable firmware and hard-coded conversion tables, and with the sources split between a 2027 and a 2035 cut-off, infrastructure teams would be wise to assume the earlier deadline.