The finding
The North Atlantic Council said in May 2024 that Russia was conducting hybrid activities against Allies, and named seven affected members. That is the alliance’s own framing and it is the only collective attribution on this flank. Everything else that names a foreign service is national, judicial, and late.
Thirty-one recorded cases sit behind this module’s hybrid layer. Eleven reach an official act naming a foreign state or service. Eight reach a court verdict — seven convictions and one prosecution that ended in acquittals.
Those are two different populations, and the gap between them is the finding. Only four cases reach both: the Vilnius arson, the Stuttgart parcel plot and two Estonian cases. Seven attributions rest on no judgment at all, because they are government acts, sanctions and findings rather than court decisions. And two convictions are recorded as naming no foreign service whatever — the Prague bus-depot arson and the arson at the Museum of the Occupation of Latvia.
A court establishes who lit the fire. A government establishes who asked for it. They are different instruments, they run at different speeds, and most cases on this flank get only one of them.
That is the finding. It is not a criticism of prosecutors, who are doing something harder and more consequential than issuing a statement: they are building a case that will survive a defence. It is a description of what each instrument can and cannot do. A parcel that ignites at a courier hub in July 2024 is charged in September 2025 and reaches trial in April 2026, where it still has no verdict. By the time the act that names the service arrives, the campaign it belongs to has moved on to a different method.
What the range shows
The median hides two ends that matter more than it does.
Four days. A charge detonated under the Warsaw–Lublin railway on the evening of 15 November 2025. A metal device was found fixed to the rail near Gołąb the next day. On 19 November Polish prosecutors charged two Ukrainian citizens in absentia with sabotage of a terrorist character on behalf of Russian intelligence. Both had already crossed into Belarus. A European arrest warrant was sought on 1 December and international notices were issued.
Four days from the act, three from the discovery. What made that possible was not speed of investigation but the nature of the case: a charge in absentia against identified suspects who have left the jurisdiction does not have to survive a trial, because no trial can take place. The fastest attribution in the module is also the one least likely to be tested.
Nearly two years, and then a partial outcome. Three men were arrested in Germany and Switzerland in May 2025 over tracked test parcels that the federal prosecutor said were preparation for a sabotage operation ordered by a Russian intelligence service. On 18 August 2026 the Higher Regional Court in Stuttgart convicted one of espionage and conspiracy to commit arson, and acquitted the other two.
That is the case the whole argument turns on. A prosecutor’s attribution reached a court, and two thirds of it did not survive the hearing. An attribution made at the point of charge is a claim; only the judgment tells you which part of it was evidence.
Three cases, three different things the word “attribution” means
The cases are not on a single scale, and treating them as one is the commonest error made about this material.
Estonia is the high-water mark. In July 2025 Harju County Court convicted two Moldovan nationals over arson attacks in January 2024 on a Tallinn restaurant and a store in Osula, and found that the principal had been tasked with the operation by Russian military intelligence. That is a court finding of tasking — not a prosecutor’s assertion, not a minister’s characterisation. On flank territory it has happened here and, on the prosecutors’ wording, in Lithuania.
Poland is a charge, and says so. Five people were indicted in April 2026 over the 2024 Warsaw arsons, charged with acting on behalf of Russian intelligence. No verdict exists. In the Marywilska 44 case the direct perpetrators have not been identified at all.
Prague is a conviction with no attribution. In June 2025 Prague Municipal Court sentenced a Colombian national to eight years for setting fire to three buses at a city depot and for preparing a second attack; reporting of the judgment describes recruitment through a messaging application for a promised payment. No judicial finding names a foreign service. The Czech prime minister had said publicly that the attack was probably part of Russian hybrid activity against Czechia — which is a head of government’s characterisation of a campaign, and not a finding about this fire.
A reader who collapses those three into “Russian sabotage in Europe” has lost the distinction that the last four years of court records exist to establish.
Why this is a flank problem and not a crime-statistics problem
The military layer of this module does not answer any of it. A multinational brigade deters an armed attack. It does not prevent an arson at a shopping centre, a device in a courier network, or a charge under a railway carrying traffic towards Ukraine. The institutions that answer these acts are national police forces, national prosecutors and national courts — civil bodies, outside NATO’s command, working to evidentiary standards that were designed for prosecuting crimes and not for signalling to a state.
The result is a mismatch of clocks. Alliance statements can be made in days and carry no evidentiary weight against any individual act. Judicial findings carry decisive weight and arrive in months or years. There is nothing in between, and the flank’s hybrid pressure lives precisely in between.
That is also why this module counts damage to Baltic cables and pipelines independently of attribution. Fifteen named assets have been damaged on nine occurrence days since September 2022. Not one of those cases has reached a state attribution — including the one in which three ship’s officers have been charged, a district court dismissed for want of jurisdiction, an appeal court restored the case in August 2026, and the merits have still not been heard. One case was closed by Swedish prosecutors as an accident. Counting the damage and counting the sabotage would produce two different numbers, and only one of them can be established.
What this Note does not claim
It does not claim that attribution is pointless. The convictions are real, the sentences are real, and a court finding of tasking by a foreign military intelligence service is among the strongest public statements a democratic state can make about another state’s conduct.
It does not claim that states are unwilling to attribute. Poland attributed in four days.
It does not claim that faster attribution would reduce the activity. Nothing in the record supports that, and the Polish case is evidence against it.
And it does not claim the count is rising. More recorded cases may mean more activity, or more detection, or more willingness by states that previously handled such cases quietly to publish them — and this module has no way to separate those three. That is why no Tempo series counts attributed sabotage: about two a year with a median lag near fourteen months is a case list, not a line on a chart, and a line would fall every time prosecution slowed.
What would change the assessment
- A second collective statement by the North Atlantic Council, especially one naming a specific incident rather than a pattern. The May 2024 statement is still the only one.
- A verdict in one of the open cases — the Vilnius parcel trial above all, which has been in hearing since April 2026 and would be the first verdict on a network rather than on an individual.
- Any state attribution in an undersea case. There has never been one.
- An incident of a materially new form. Arson, parcel devices and railway sabotage are different capabilities, and the form changing has so far told this module more than the count changing.
- A national legal or institutional measure that shortens the interval — a change in how these cases are prosecuted, rather than in how many there are.
The interval is the observable. It is measured case by case from published dates, on the cases that have a single act date to measure from; it currently sits near fourteen months at the median, and it is the one number on this flank that would mean something different if it moved.