GPS interference reached Los Angeles. The advisory system is built for the tests, not the surprises

GPS interference over a major North American metropolitan area is now a reported operating condition. On 25 September 2026, United flight 1962, an Airbus A321neo, reported navigation problems at 07:36 UTC shortly after departing Los Angeles International. United flight 1866, cruising at 35,000 feet between San Francisco and Dallas/Fort Worth, reported interference at 08:15 UTC. ^[1] An Alaska Airlines Boeing 737 MAX 9 out of San Diego was affected in the offshore sector west of Los Angeles. ^[2] No aircraft lost control; all retained inertial reference and ground-based navigation. No authority has identified the source. ^[1][2]

That last sentence is the finding. The interference reached air traffic control from three cockpits, and no mechanism exists that will name what produced it. That is the design, not a failure: the civil apparatus for radio-frequency interference in aviation is a scheduling and reporting instrument, built around interference declared in advance.

The interference that is advertised

Scheduled GPS interference testing in the United States is published as flight advisories with a predicted degradation envelope. One issued in September 2026 for a test area in the south-eastern United States is representative. Designated CSG4 GPS 26-04 Revision 3, it covers windows on 4, 7 and 9 September, centred 41 NM from the Charleston VOR on the 164-degree radial, and gives the radius of possible GPS unavailability by altitude: 295 NM at FL400 and above, 247 NM at FL250, 178 NM at 10,000 feet, 132 NM at 4,000 feet AGL, and 66 NM at 50 feet AGL. Expected effects include WAAS, GBAS and ADS-B. ^[3]

The altitude table is the part practitioners outside aviation consistently miss. Reach is a line-of-sight problem, so the affected volume is a cone, not a circle on a chart. A test whose ground footprint is a county affects an airway four hundred kilometres away at cruise. The same advisory tells operators to monitor NOTAMs, gives a 24-hour minimum for changes, and points to the Aeronautical Information Manual for anomaly reporting. ^[3]

Southern California carried four scheduled interference notices in September 2026: Twentynine Palms from 14 to 19 September, the Point Mugu Sea Range from 15 to 24 September, Edwards Air Force Base from 16 to 18 September, and China Lake from 20 to 25 September. The regional users’ group that consolidates them advises that notices are published at least 72 hours ahead, and records that testing can degrade the signal from ground level upward and well beyond the test site’s own area. ^[4] That 72-hour figure and the advisory’s 24-hour minimum differ; the shorter one is the floor.

The China Lake window ran to 25 September, the date of the Los Angeles reports. Nothing in the record connects them. The point is narrower: an operator holding both the notice list and the pilot reports cannot tell from either document whether the two are the same event. The scheduling instrument was never built to answer that.

The interference that is not

In the same month, two aircraft reported GPS loss in Swedish airspace. A KLM service from Amsterdam to Helsinki on 17 September reported persistent signal loss and requested an avionics reset in flight. An Air Canada service from Delhi to Toronto reported the same class of interference the following day. The affected corridor runs around Gotland and Oland and along routes towards Finland, Poland and the Baltic states. ^[5]

The scale figures come from one account. Jorgen Andersson of the Swedish Transport Agency is reported as saying interference has remained at a high level with no clear decline since late winter, and that spoofing has become more systematic and more geographically widespread. The same account gives 142 reports logged by 28 May 2026 against roughly 336 for all of 2025, and cites an ICAO filing for January to April 2025 covering approximately 123,000 affected flights across 365 airlines. ^[5] The filing itself was not available to this desk.

Regulators have moved on guidance rather than attribution. The FAA’s GPS and GNSS Interference Resource Guide reached version 1.1, flagged to operators in March 2026, covering trends, aircraft effects, pilot procedures and training. ^[6] A later account dates the original to December 2025 and reports its figures: GPS signal loss per 1,000 flights up 65 per cent in the first half of 2024 against the same period in 2023, and Cyprus logging more than 5,600 spoofing incidents in two months. Its hotspots include the eastern Mediterranean, the Black Sea, the Russia and Baltic region, the India-Pakistan border, Iraq and Iran, the Korean peninsula and the area around Beijing. ^[7] For operators out of Kuala Lumpur, the nearest persistent cases on that list are the Korean peninsula and Beijing.

EASA published revision 4 of Safety Information Bulletin 2022-02 on 3 July 2026, the first update since July 2024. ^[8] It advises operators to verify position using alternative navigation aids near affected areas, and notes that Type B electronic flight bag applications showing near-real-time interference data allow better anticipation by crews. ^[9] EASA and EUROCONTROL have also published a joint action plan pooling monitoring data across Europe; the same record recalls the 2022 Denver case, in which an unauthorised transmitter degraded GNSS reception for civil flights and air traffic control. ^[10]

Why discrimination is the hard part

Three causes produce the same cockpit symptom. A scheduled military test, an unauthorised or faulty transmitter, and a hostile emitter all present as satellite count collapse, integrity flags, position drift and autopilot misbehaviour. Practitioner guidance teaches the symptom set rather than the cause, treating simultaneity across systems as the signal worth trusting. It keeps the distinction that matters operationally: jamming floods the band so the receiver cannot hear the satellites, while spoofing transmits a plausible signal so the receiver reports a confident wrong position. ^[11]

Attribution requires what the aviation system does not routinely collect: bearing, time of arrival and signal characteristics from multiple receivers, correlated. Reports arrive as free text after landing. The multilateral response reflects this. The heads of ITU, ICAO and IMO have jointly called on states to protect the radionavigation satellite service, strengthen resilience, retain conventional aids as backup, coordinate across regulatory, aviation, maritime, defence and enforcement authorities, and establish comprehensive mechanisms for reporting interference. ^[12] Four of those five are mitigation. The fifth asks states to build the reporting capability that would make attribution possible.

What to do

• Treat the scheduled test feed as an operational input. Advisories, NOTAMs and users’ group consolidations are the only source distinguishing a known emitter from an unknown one, and only if read before the event. ^[3][4]
• Plan against the altitude envelope, not the test site. A footprint of 295 NM at FL400 and 66 NM near the surface is one emitter; route exposure has to be assessed at the altitudes actually flown. ^[3]
• File the report even when the cause looks obvious, and capture what a post-flight narrative cannot reconstruct: time, position, altitude, which systems flagged, what recovered first, whether a reset was needed. Reporting is the only mechanism that turns individual events into a dataset. ^[6][11]
• Keep conventional navigation currency live. Every regulator here lists ground-based aids and inertial reference as the backup, which is a training commitment, not a line in a bulletin. ^[9][12]

The ELINT read

Electronic intelligence begins with discrimination, and civil aviation has arrived at that problem from the other direction. It has a mature process for interference that is declared and almost none for interference that is not. The Los Angeles reports are unremarkable individually and instructive together: three crews, one metropolitan area, forty minutes, and a record that establishes what happened but not what caused it. Where scheduled testing and unexplained events share airspace in the same week, the calendar becomes a collection requirement. An operator that can say which emitters were advertised has separated the signal it can plan around from the one it cannot. That separation is the whole of the analytical work available today.