DKDKCISSP
Software Defect — Critical Service Outage

NATS: a one-millisecond timing window stopped UK air traffic for most of a day

A valid request for an aircraft code collided with a higher-priority message inside a legacy software module. Flight data was corrupted, the control centre lost its link to the national system, and UK departures were stopped for about four and a half hours. It was not a cyber attack. It was a latent defect nobody knew about.

Trigger window
About 1 millisecond
Time to declare a major incident
2h 32m
UK departure stop
About 4.5 hours
Flights delayed, cancelled or diverted
More than 2,000
Estimated UK economic cost (CMC)
About £33m
Cause
Software defect, not an attack
Aircraft separation
Maintained throughout

Background

NATS runs air traffic control for UK airspace. Its National Airspace System (NAS) is the core flight data system behind many of its control tools: it keeps flight plans, processes amendments, hands out the identification codes (squawk codes) that let controllers match an aircraft to its flight plan, and links flight plans to radar tracks. London Area Control (LAC) handles flights mainly above 24,500 feet over England and Wales, around 6,500 flights on a typical September day.

This was not the first major NATS failure. In August 2023 a flight plan failure grounded flights across the UK, and the review that followed produced recommendations on incident management. NATS states that the September 2026 incident is unrelated to that failure and to a radar issue in July 2025.

From one request to a UK-wide restriction

Valid squawk code request, 10:00Paused by a higher-priority messageLatent defect corrupts flight dataExposure window of about one millisecondLink to the national system dropsFirst for 45 seconds at 10:02, then for good at 13:32Manual fallback and UK-wide restrictionsDepartures stopped, flights divertedControlled restart and data clean-upRestrictions fully lifted at 19:30
Fig. 1 — The trigger was ordinary. Only the timing was rare.

What actually happened

At 10:00 on Tuesday 8 September 2026, a controller made a correct manual request to allocate a squawk code to a valid flight plan. While it was being processed, the NAS received a message for a higher-priority task, so the request was paused. When it resumed, a legacy software defect meant it did not resume correctly and the output was corrupted. Some later flight data updates were affected. The window in which this can happen is about one millisecond: a higher-priority request had to arrive in that exact moment, part-way through updating a value.

At 10:02 controllers and engineers saw a notification that the link between the LAC system and the NAS had dropped. It cleared after 45 seconds, and at 10:06 a standard engineering incident was raised noting the system had recovered. Health checks found no hardware fault. At 12:32 the link began dropping repeatedly, which was the first sign controllers saw of sustained trouble, and NATS declared a major incident. Restrictions were applied from 12:45, and at 13:32 the link dropped and did not come back.

Controllers moved to practised fallback procedures, including manually handing aircraft over to neighbouring air traffic units. Restrictions were placed across UK airspace so the NAS could be restarted and flight data reloaded. The restart ran from 15:17 to 16:09, followed by reconciling duplicate flight plans and code mismatches until 18:50. All airspace restrictions were lifted at 19:30. At no point did aircraft lose safe separation.

NATS handled 6,094 flights that day against a forecast of around 8,000, and more than 2,000 flights were delayed, cancelled or diverted. On 4 November the Cyber Monitoring Centre estimated the cost to the UK economy at around £33m, almost all of it falling on airlines and other organisations rather than on NATS.

Still open: the final reportNATS published a preliminary report on 16 September and said a full Major Incident Investigation would follow within 60 days. We have not seen that final report, so the precise millisecond timing, any other conditions that raised the odds of the defect firing, and the review of NATS's command and control are all still outstanding.

Root cause

Confirmed
  • A legacy, previously unknown software defect in the NAS module that processes requests to allocate or reassign squawk codes.
  • The trigger was a valid, correctly made manual request, not an abnormal flight plan or operator error.
  • A higher-priority message paused the request; when it resumed, the defect meant the update did not resume correctly and corrupted the data.
  • The exposure window was about one millisecond, which is why it had not happened before.
  • The link between the LAC system and the NAS dropped by design when processing corrupted data kept timing out, to protect the integrity of both systems.
  • NATS found no evidence of a malicious actor or cyber activity, and says it was not caused by incorrect actions by military or civil operators.
Reasonable hypothesis — not confirmed
  • NATS says it is still investigating whether other conditions in the wider system raised the probability of the defect being triggered. That has not been reported.
  • How long the defect had been in the code, and why testing and assurance never reached it, has not been disclosed.

Timeline

A correct manual squawk request is paused by a higher-priority message and corrupts flight data when it resumes. Nobody knows yet.
Controllers and engineers see the link to the NAS drop. It returns after 45 seconds.
An engineering incident is raised as recovered and stable. Health checks find no hardware fault, and specialists begin looking at timeout conditions.
The link starts dropping repeatedly. Some automation is lost. Bronze command is activated at 12:35.
The first airspace restrictions and a stop on UK departures begin. This is when airlines and passengers feel it.
The link drops and does not recover. Controllers switch to manual coordination and stricter limits follow at 13:38.
The NAS is restarted and flight data is reloaded from the LAC system.
Duplicates and code mismatches are cleared by 18:50. All restrictions are lifted at 19:30.
Fig. 2 — Two and a half hours passed between the first sign and a declared major incident.

Detection

How it was discovered

Internal monitoring and controllers. The first link-drop notice at 10:02 cleared on its own; the sustained link instability that began at 12:32 was the first indication to controllers that something was wrong. NATS engineers then traced it to flight data processing in the NAS.

Time to detect

About 2 hours 32 minutes from the corrupted data at 10:00 to a declared major incident at 12:32

Warning signs that were missed
  • A link drop at 10:02 that lasted 45 seconds and recovered by itself was logged as a routine engineering incident. In hindsight it was the first symptom of corrupted data.
  • Health checks between 10:06 and 12:32 looked at hardware and found nothing, while the real fault was in data. The report says there were no indications of corruption in that period.

What to monitor for in your own environment

A link between two critical systems dropping and timing out on a specific piece of data, even once, even if it self-heals

System health and link-status logs

A processing request that exceeds its normal time and then times out repeatedly on the same data

Application timing and timeout metrics

Output from a priority-interrupted task that fails a data-integrity check before it is stored

Application-level validation and integrity checks

Correlation between flight-data updates and link instability, which the engineers started on at 12:53

Event correlation across system and data logs

Containment, eradication & recovery

Containment

12:32 – 15:17
  • Activated Bronze (12:35), Silver (12:49) and Gold (13:08) command teams under the standard incident process.
  • Applied sector limits, stopped UK departures and limited arrivals, and told airlines and airports through the EUROCONTROL portal.
  • Switched controllers to practised fallback procedures with manual hand-overs to neighbouring units.
  • Contacted the software supplier for specialist support at 12:53, and held the first stakeholder coordination call at 13:40.

Eradication

13:45 – 16:09
  • Concluded at 13:45 that the NAS was the likely cause and that a controlled restart and reload was the best way to recover.
  • Reviewed and safety-approved the restart plan, then restarted the NAS and reloaded flight data between 15:17 and 16:09.
  • A permanent software fix was developed by the supplier and delivered, and was under safety testing when the preliminary report was published.

Recovery

16:09 – Sep 11
  • Reconciled data that had gone out of step: duplicate flight plans, aircraft code and callsign mismatches, and plans filed during the restrictions. This finished at 18:50.
  • Lifted restrictions step by step, with all of them removed at 19:30, and ran normal capacity the next day.
  • Held 15 stakeholder calls in total, with the last on Friday 11 September.
  • Put an extra engineering process in place for link-drop events, with defined escalation, and kept Gold command active after the event.

Security control analysis

Held
Fallback procedures and annual controller training

Controllers followed practised fallback procedures. The 2025/26 refresher syllabus included a link failure between the LAC system and the NAS. Aircraft stayed safely separated.

Held
Automatic link drop to protect system integrity

The link dropped as designed when corrupted data kept timing out, which kept the fault inside one control centre's system.

Failed
Software assurance and testing of legacy code

A defect in legacy code with a one-millisecond window had never been found. How long it had been there, and why testing missed it, is not disclosed.

Partial
Detection and correlation of the first link drop

The drop at 10:02 was seen and logged, but treated as a recovered routine incident. The link between drops and data corruption was not made until the failures became frequent.

Held
Incident management after the 2023 failure

A dedicated Major Incident Manager, an ATICCC briefing cell and Bronze, Silver and Gold command all ran as designed, and NATS says earlier recommendations were implemented and independently verified.

Partial
Speed of restart and data reconciliation

A controlled restart was safe and documented, but wider restarts affect multiple units, and the 18:00 call said processing was returning to normal more slowly than expected.

Unknown
Capacity planning for a multi-hour restriction

The Cyber Monitoring Centre recommended reviewing how well NATS, airlines and airports coordinated and whether earlier outage-duration estimates would have limited disruption. NATS's own review of this is pending.

Controls that would have prevented or limited this

Fix
Add integrity checks to the output of any task that can be interrupted by a higher-priority message

The corrupted squawk allocation would have been caught and rejected at 10:00, before it reached later flight data updates.

Closes: Software assurance and testing of legacy codeCIS Control 16 — Application Software Security
Fix
Test concurrency and timing: inject priority interrupts at many points inside critical update routines

A race condition with a one-millisecond window is the kind of fault timing-focused testing is built to find.

Closes: Software assurance and testing of legacy codeCIS Control 16 — Application Software Security
Fix
Alert on any link drop between two critical systems that repeats or points to the same data, even if it recovers

The 10:02 drop would have triggered a data-focused investigation 2.5 hours earlier.

Closes: Detection and correlation of the first link dropCIS Control 13 — Network Monitoring and Defense
Fix
Rehearse the full restart-and-reconcile path, not just the fallback, and publish expected recovery times to partners

Recovery would have been faster, and airlines could have planned around a realistic restriction window.

Closes: Speed of restart and data reconciliationCIS Control 17 — Incident Response Management
Fix
Keep tested, recent recovery procedures for core data systems and measure reload time

The restart from 15:17 to 16:09 and the data clean-up to 18:50 become predictable instead of discovered live.

Closes: Speed of restart and data reconciliationCIS Control 11 — Data Recovery

Blast radius

Who felt it

  • More than 2,000 flights delayed, cancelled or diverted, against a forecast of around 8,000 flights
  • Hundreds of thousands of travellers, with about 450,000 affected by cancellations or delays of more than two hours, according to the Cyber Monitoring Centre
  • Airlines, airports, ground operations and tour operators across the UK, with 70 domestic organisations counted in the CMC estimate
  • UK departures stopped for around four and a half hours across a six-hour period

What did not fail

  • Aircraft separation and safety. NATS says no aircraft lost safe separation at any time.
  • Radar. Controllers could speak with aircraft and monitor them on radar throughout.
  • The link between systems, which dropped by design to contain the fault
Fig. 3 — Almost all of the cost fell on airlines and other organisations, not on the operator.

Hours from the first corrupted data to full recovery

Corrupted data → major incident declared2h 32mMajor incident → restart begins2h 45mRestart → data reconciled3h 33mFirst corrupted data → all restrictions lifted9h 30m
Fig. 4 — The longest stretch was not diagnosis. It was getting a restarted system back into step.

What they got right

Controllers followed practised fallback procedures and kept every aircraft safely separated for the whole incident.

The system was designed to drop its link when it could not process data, which limited the fault to one control centre's connection.

Command roles, the major incident manager and a single stakeholder briefing cell all ran as designed after the 2023 failure, and all stakeholders got the same information at the same time.

NATS published a preliminary report within eight days, with a timeline down to the minute, and said plainly that the cause was a defect in its own software.

The permanent fix was developed quickly by the supplier, and extra monitoring and escalation were put in place while it was tested.

Why this keeps happening

Legacy code is trusted because it has run for years without incident. A defect with a one-millisecond window can sit unseen for as long as that timing never happens.

The earliest signal looked harmless. A self-healing blip on a link is easy to classify as routine, so it is routinely classified that way.

Safety-critical systems are built to fail safe, and failing safe often means stopping. The cost of stopping then falls on everyone downstream.

Lessons for your own environment

  1. A self-healing fault is still a fault

    A 45-second drop that fixed itself was the first sign of the incident. Treat the first recovery as a clue, and ask what data was in flight.

  2. Checks that look at hardware cannot find a data fault

    Between 10:06 and 12:32 engineers found nothing wrong with the infrastructure because nothing was. Pair health checks with integrity checks on the data.

  3. Legacy code needs timing tests, not only functional tests

    The defect needed an interrupt in a one-millisecond window. Race conditions are found by deliberately interrupting code, not by running it normally.

  4. Fallback training worked, so train recovery as well

    The manual procedures held. The slow part was restarting and reconciling data across connected systems, which is worth rehearsing just as hard.

  5. Partners need an honest estimate of how long it will last

    The Cyber Monitoring Centre asks whether earlier, more frequent duration estimates would have helped airlines limit disruption. Plan for a multi-hour restriction.

  6. A resilience cost is paid by customers first

    Almost all of the £33m fell on airlines and others, not NATS. Whoever depends on your service should know how it behaves when it stops.

CISO & board lessons

  • Ask which of your most critical systems rely on legacy code that nobody has tested for timing and concurrency faults, and what the plan is for each.
  • Check whether your monitoring correlates a brief, self-healing link drop with the data being processed at that moment, not only with infrastructure health.
  • Treat non-malicious failures as incidents of the same weight as attacks. A software defect stopped a national service for most of a day, with no attacker involved.
  • Rehearse recovery end to end, including restart and data reconciliation across connected systems, and record how long each step takes.
  • Make sure the people who depend on you hear expected recovery times early, even when those times are uncertain.

Action plan

First 24 hours

  • Move to documented fallback procedures and limit load to what manual operation can handle.
  • Activate command roles and open a single briefing channel so every partner hears the same thing at the same time.
  • Contact the supplier that maintains the affected software and preserve logs and data from before the first symptom.

First 7 days

  • Restart and reconcile in a controlled sequence, and bring automation back gradually as confidence grows.
  • Put an interim monitoring and escalation process in place for the specific failure signature while a permanent fix is tested.
  • Publish a preliminary account with a timeline and the confirmed cause.

First 30 days

  • Complete a root-cause review, including whether other conditions raised the chance of the defect firing.
  • Deploy the safety-tested fix, and re-run timing and concurrency tests on similar modules.
  • Review how partners, airlines and airports coordinated, and publish recovery-time expectations.

Could this happen in your organization?

  • Which of your critical systems run legacy code that has never been tested for timing and priority-interrupt faults?
  • If a link between two critical systems drops for 45 seconds and recovers, does anyone ask what data was in flight?
  • Do your health checks cover data integrity, or only hardware and connectivity?
  • Have you rehearsed restarting your core data system and reconciling it with connected systems, and do you know how long it takes?
  • Do your partners know how long a restriction could last, and have they planned for several hours?
  • Would you treat a non-malicious software failure with the same incident process as an attack?

If this happened to you: what to ask, and what to say

Knowing the lesson isn’t the same as knowing what to do with it Monday morning. A security lead, a CISO, and a customer-facing team each need something different from this incident.

  • “What happens if a higher-priority message interrupts our most important update routine at the worst possible moment?”
    If nobody has tested it, assume nobody knows.
  • “Do we alert on a link that recovers by itself?”
    The 10:02 drop was the first and cheapest warning of the whole incident.
  • “How long would a controlled restart and data reconcile take for us, and when did we last time it?”
    NATS's restart was 52 minutes, followed by nearly three more hours to put data back in step.
  • “Who tells partners how long this will last, and when?”
    Pick the owner before the incident, not during it.
  • State the cause without calling it a cyber attack.
    It was a software defect, and it is still a board-level resilience issue.
    “A rare timing fault in an old part of our flight data software corrupted data. The system dropped its connection by design, we moved to manual procedures and restricted traffic to stay safe, then restarted and reconciled the data.”
  • Separate confirmed findings from what is still open.
    The final report is outstanding.
    “The cause is confirmed. How long the defect had existed and whether other conditions made it more likely are still being investigated, and our final report will say.”
  • Show that the fix is broader than one module.
    Boards want to know similar code has been checked.
    “We have a permanent fix in safety testing, interim monitoring in place, and a review of similar legacy modules and our recovery procedures.”
  • Be clear about safety first.
    Say it plainly, then describe the disruption.
    “At no point was safety in question. Controllers kept every aircraft separated while we restricted traffic to keep it that way.”
  • Give a realistic view of recovery.
    Partners can plan around an honest range.
    “Restrictions were in place for about six hours, and clearing the backlog took more than two days. In future we will give estimates of duration earlier and update them more often.”
  • Own the failure.
    Apologise for the impact, not for the weather.
    “This was a failure of our own software and we apologise to everyone affected. Here is what has changed so that it cannot recur in the same way.”

References & confidence

  • ConfirmedNATS — Major Incident Preliminary Investigation Report, NAS incident of 8 September 2026 (issued 16 September 2026) — Primary source for the timeline, technical cause, response and next steps.
  • ConfirmedNATS — NATS publishes preliminary report on technical incident of 8 September (18 September 2026)
  • ReportedInfosecurity Magazine — NATS Air Traffic Control Failure Cost UK Economy £33m (5 November 2026) — Summarises the Cyber Monitoring Centre report published 4 November. The £33m figure is the CMC's modelled estimate, not a NATS or government number.
  • ReportedCyber Monitoring Centre — report on the NATS incident (4 November 2026) — Read through press coverage; we did not read the report itself.
  • UnknownNATS — final Major Incident Investigation report — Due within 60 days of the incident. Not reviewed for this study.
  • UnknownCivil Aviation Authority — independent review of the incident — Reported elsewhere as requested by government; its status and findings were not confirmed from a primary source.
This case study is built entirely from public reporting on this incident — see References & confidence above for what's confirmed, reported, or still unconfirmed. Nothing here is an estimate or an illustrative placeholder.