The probability remains small. But the trend matters. The Forecasting Machine (TFM) has raised to 10% the probability that a direct military confrontation between Russia and NATO will occur. A week ago it stood at 9.5%. In our latest briefing it was 9.3%. The update of September 15 therefore implies a rise of seven tenths of a percentage point in just a few hours. This does not mean, however, that a war between Russia and NATO is likely. For now, nine out of ten scenarios point in the opposite direction. But something is happening beneath that percentage. In less than 24 hours there have been two episodes that, while not constituting the unequivocal event TFM is trying to forecast, come unusually close to its limits.
“The Forecasting Machine (TFM) has elevated to 10% the probability that a direct military confrontation between Russia and NATO will occur”
The first occurred over Lithuania. A drone penetrated its airspace from Belarus and was downed by Italian fighter jets deployed at Šiauliai as part of the NATO air defense mission. Subsequently, engineers found and neutralized an explosive among its wreckage. The presence of a charge makes the episode considerably more serious than initially thought. But the fundamental question remains unresolved: who was operating the aircraft?
The second occurred in the Baltic. A Russian guided-munition ship fired two flares at a Danish military helicopter conducting surveillance in international waters. According to Danish authorities, one flare passed within a close distance of the helicopter. Copenhagen summoned the Russian ambassador but chose not to request consultations within NATO because it does not consider what happened a direct attack on its territory. Moscow responded by accusing the helicopter of performing dangerous maneuvers.
Neither of the two events allows us to state that Russia and NATO have entered into a direct military confrontation. But both compel us to ask an uncomfortable question. What would have to happen exactly for us to say that that confrontation has begun?
The 10% Problem
Every forecast needs a question. And every question needs a definition. Here lies one of the more interesting issues in the monitoring that Agenda Pública is conducting of The Forecasting Machine. The public rule used by the forecast speaks of a “direct military confrontation” confirmed by reliable sources. However, the practical evolution of the analysis seems to demand a higher threshold: some form of deliberate combat, casualties, or political and military acknowledgment that an attack has occurred. The events in Lithuania and Denmark demonstrate how much that difference matters.
“Fighters from an NATO country used their weapons to destroy an aircraft that had penetrated the airspace of another alliance member”
In Lithuania there was an unequivocal military action. Fighters from an NATO country used their weapons to destroy an aircraft that had penetrated the airspace of another alliance member. We also know that it was carrying an explosive. But we still do not know who fired it. In Denmark the opposite happens exactly. Here we know both actors. A Russian frigate fired flares at a Danish military helicopter, but there was no returning fire, no casualties and no damage, and Copenhagen decided to manage the episode diplomatically.
We therefore have the two halves of a possible confrontation without either one alone completing the definition. Lithuania has action, but no attribution. Denmark has attribution, but no combat. That ambiguity does not invalidate the forecast; it makes it more interesting because it probably tells us something about the nature of European security in 2026: the boundary between peace and war is becoming much harder to identify.
Florence Gaub and Scenarios B and C
Just a few hours before these events, the scenario expert Florence Gaub had posed precisely this problem in Warsaw. The Director of the Research Division at the NATO Defense College spoke at the Europe Future Forum, organized by Visegrad Insight and in which Agenda Pública participated, and issued a warning particularly relevant for interpreting the evolution of the forecast.
Europe must continue preparing for the Scenario A: a Russian conventional attack on Poland or one of the Baltic states that provokes a military response from NATO. But it would be a mistake to assume that Russia will necessarily act according to the scenario we have spent years rehearsing. Gaub argued for the need to imagine the Plans B and C. We can count tanks, aircraft, soldiers, ammunition and brigades. We can calculate how long it would take Russia to rebuild its forces after Ukraine. We can study how Europe must expand its industrial output and reduce its dependence on the so-called ‘American enablers’. But we do not know what the next war will look like; history shows we are surprisingly poor at predicting it.
“But we do not know what the next war will look like; history shows we are surprisingly poor at predicting it”
That is why Gaub revisited in Warsaw the work of Lawrence Freedman on the history of military predictions. Societies tend to imagine the next conflict using the categories of the previous one. And then reality finds another door. Perhaps that is precisely what we are seeing now.
The War That Does Not Seem Like a War
For years we have used the term “hybrid war” to describe what happens between peace and conventional war. Gaub believes that even that expression has ceased to be particularly useful. Under that umbrella coexist forms of pressure that are very different: sabotage, cyberattacks, drones, electronic interference, information operations, incidents against critical infrastructure and covert actions. The question is no longer merely whether Russia could someday prepare to attack NATO militarily, but that it can inflict damage today without clearly crossing the threshold that would force the Alliance to respond militarily. That certainly changes the reading of the forecast. If our sole question is whether we will see , we are probably looking at the wrong indicator.
The most immediate risk may lie precisely in the accumulation of small events: a drone enters allied territory and a fighter shoots it down; a frigate fires flares and a helicopter continues its mission; a submarine cable is damaged; an airport halts operations; a radar misreads a trajectory. Each isolated event may be manageable, the problem arises when two decisions taken in seconds turn an ambiguous incident into a sequence of action and reaction. That is the mechanism we are tracing at Agenda Futura.
The Window of Vulnerability
The second element of Gaub’s analysis also helps interpret the 10% of TFM. Europe is going through a “vulnerability window”. The next five years will be particularly delicate because Europeans are trying to build the capabilities necessary to assume a much larger share of their own defense while still relying heavily on the United States. Washington has spent years making clear that its main strategic challenge is China, while for Europeans the immediate threat remains Russia.
The divergence forces Europe to do more, but five years is a very short time to transform the armed forces. Gaub recalled that the United States needed roughly a decade to reform its military structure after Vietnam. Europe is now attempting to reshape the capabilities of more than thirty states, with different governments, industries, budgets and electoral cycles.
“While Russia maintains a substantial portion of its forces committed to Ukraine, its ability to open a simultaneous conventional conflict against NATO is lower”
Ucrania provides time. While Russia keeps a substantial portion of its forces committed there, its ability to launch a simultaneous conventional conflict against NATO is diminished. Europe has a few years to close part of its military deficit. But that does not mean Russia has to wait; it can operate below the threshold. And that is where the Baltic incidents take on another dimension.
The Baltic as a Laboratory
The two incidents of the past hours point to the same region. The Baltic Sea is becoming the principal European laboratory for this new security regime. In a relatively small space, Russian and allied forces, Kaliningrad, the borders of Poland and the Baltic states, important sea routes, energy and telecommunications infrastructures, military bases and a growing density of air and naval operations coexist.
The downing of the Lithuanian incident also provides a reassuring signal. The defensive architecture worked: the object was detected, the Italian fighters took off from Šiauliai and the threat was neutralized. The Lithuanian Defence Minister later explained that national authorities participated in the decision-making from the very beginning and that the operation lasted about thirty minutes. It is precisely the kind of allied cohesion that increases deterrence.
Denmark provides another counterweight. Copenhagen could have raised the political temperature but chose not to. It summoned the Russian ambassador and kept the dispute on the diplomatic plane. Moscow responded with a diplomatic protest as well. The same day yields arguments in both directions. The number and severity of incidents rises. But the mechanisms designed to prevent those incidents from escalating also work. That is why the 10% is plausible as a description of an uncertainty, but it would be a mistake to read it as an indicator that war is inevitably approaching.
The Danger of Signaling 2030
Gaub introduced another relevant warning in Warsaw. He has tallied more than fifteen NATO officials or allied country representatives who have publicly spoken of a Russian attack before 2030. He understands the political objective: to create a sense of urgency to accelerate European rearmament. But there is a risk because fear does not always mobilize; it can also paralyze. And above all, no one knows that Russia will attack in 2030. Confusing preparation with prediction would be a mistake.
“Governments’ obligation is to prepare for low-probability but high-consequence scenarios, not to present those scenarios as inevitable events”
The obligation of governments is to prepare for scenarios of low probability but extraordinary consequences, not to present those scenarios as inevitable events. That is precisely what makes a tool like The Forecasting Machine useful if interpreted correctly. A 10% does not mean “Russia will attack”; it simply means there is enough evidence not to treat the scenario as negligible.
What We Have to Watch Now
The first variable is Lithuania. The forensic investigation will have to determine the drone model, its navigation system, the exact nature of the explosive, and, most importantly, who operated it. Until then, attributing it to Russia would go beyond the evidence available. The Lithuanian authorities themselves have explicitly asked not to rush to conclusions. The second is Denmark. We will have to know the military assessment of the flare trajectory and distance and determine to what extent it was deliberate intimidation or a maneuver that ended up creating greater risk than expected. The third remains Ukraine. Russia launched around 200 drones during the latest wave while Russia and Ukraine continued to attack energy infrastructure despite discussions about a possible moratorium. As long as that intensity persists, the risk of spillover to neighboring countries will also endure.
And there is a fourth variable we must begin to incorporate into our radar: the distance between real war and recognized war. It may be that the event that finally triggers the forecast is not an invasion. It could be a drone whose authorship is established hours later, a ship that misinterprets a maneuver, or a misfired missile.
The Forecasting Machine places the risk today at 10%. But the events in Lithuania and Denmark leave a more important question than the percentage. For decades we knew approximately where peace ended and war began. In the Europe of 2026, that line again becomes dangerously hard to see.
