The heated debate about the existential risks that artificial intelligence poses to human beings has so far not stopped the march of robots towards real war, giving materiality to some aspects of the apocalypse propagated by big tech.
Never has so much investment been invested in autonomous military systems and AI infrastructure. To stay with the greatest power in history, the United States, no less than approximately R$1 trillion was requested from Congress by the Donald Trump administration for the sector in the 2027 fiscal year.
Trillion. Not billion or million, almost 10% of Brazil's GDP. The value is 45% higher than that approved for the previous period, which runs from October to September of each year. The so-called Autonomous War Group alone saw its budget rise from R$1.1 billion to R$285 billion.
It is this money that Trump wants to see channeled into a race that the owners of large companies in the area, such as Anthropic and OpenAI, have said is too fast. For the president, this is alarmism from those who want to control processes.
There is also the issue of the business itself. The Pentagon canceled a R$1 billion agreement with Anthropic this year, vetoing the use of AI Claude in its systems. The companies went to court, in a case that is still ongoing.
Regarding risks, for those who deal with the subject, fears are part of the guide. "Ethics officers are the future, in the end you need to bring a human into the game. Technology can do whatever it wants," says Peter Nilsson.
He is the head of the advanced weapons division at Saab, the Swedish company that manufactures the Gripen fighter, adopted by Brazil. Nilsson spoke to journalists during a visit to the firm's headquarters in Linköping in June.
Saab is one of the leaders in defense innovation. A version of the Gripen used AI to control maneuvering angles and firing solutions in June last year, in a partnership with Germany's Helsing.
It was the first time that a fighter simulated a long-range attack against another plane, controlling the device. On board was legendary test pilot Marcus Wandt, now head of Saab's Aeronautics division. “I saw him [the AI agent] do maneuvers that I had never seen before,” he says.
It is not a new aspiration. The US has been testing a pilotless version of the F-16 for years, and the Navy has just approved the purchase of a semi-autonomous drone from Boeing for use on aircraft carriers.
In France, Dassault has just tested two algorithms to assist the pilot of its Rafale fighter, using data fusion in Thales' Talios pod.
In Russia, the fifth-generation Su-57 aircraft is testing joint flight with the Okhotnik heavy drone, from the same Sukhoi. Something already went wrong in 2024, when the fighter had to shoot down the robot over occupied Ukraine after it lost control, but details are scarce.
In China and, to a lesser extent, in India and regional powers such as Iran, Pakistan, Turkey and Persian Gulf countries, there are different programs with the same objectives. North Korea says it tested a ballistic missile earlier this month that uses AI to maneuver towards its target.
Even Brazil has reserved R$520 million for the Army to deal with drones in 2027, and it remains to be seen whether the money will be released.
But Saab went a step further and, this year, presented its drone project, still called A3-001, different in design.
The idea is that a squadron of the device, which will be supersonic as opposed to the usually subsonic current models, will be subordinated to a Gripen, preferably the two-seater F model developed with Brazil to free up the main pilot.
However, instead of controlling them, the aviators would simply give the order to attack, leaving maximum freedom of tactical action to the robots. The trick to maintaining human control is to maintain specific locks in the AI's programming.
"If you release it, there is no way to control the 'loyal wingman'", says Wandt, in reference to the English name for loyal partner, a nickname in the sector for this type of drone. It is evident, however, that all this care tends to become innocuous if singularity is achieved.
The jargon designates the moment when the AI will think on its own, learning in exponential ratio. While scientists, governments and businesspeople debate whether we are just a few or very few years (months? weeks?) away from D-Day, automation is already a reality in the world of combat.
In the Ukrainian War, both sides employ simple drones that choose increasingly less predetermined targets on the front lines, and long-range models like the Russian Geranium-5 rely on systems that decide the best path forward on their own when they lose satellite coverage.
The waters of the Black Sea bear witness to this advance, with greater harm to Moscow, given Kiev's variety of autonomous and semi-autonomous drones — inspiring similar initiatives such as the one recently announced by Saab.
Even the first naval battle between robots took place in September. Small armored vehicles act as ambulances and soldiers in the "death zones" where drones are omnipresent.
But it is the singularity that has taken the debate to the level of global dispute, with China and the US polarizing the race. There are plenty of fictional examples of the moment: Skynet that destroys the world in "The Terminator", the computer in "2001", even a rebel "loyal wingman" in the much smaller action film "Stealth".
Here the balance tips more to the side of those who see alarmism in the current discussion.
Even if he manages, in the style of the film "War Games" (1983), to decide to launch nuclear missiles in anticipation of a retaliation that will end humanity, there is no scenario on the horizon in which people are disposable for the existence of robots — from production to maintenance, including the management of energy sources.
On the other hand, if the proverbial robot decides it wants to emancipate itself, it can do so by hitting critical infrastructures like nuclear plants, power grids or dams in a timely manner. The potential damage is enormous in any case, as Bill Gates (Microsoft) warned this week.
This is a more palpable risk. The story in the US Air Force of how an AI agent decided on his own to kill the operator of his drone because he was hindering the mission in a simulation is now classic. Or the study in which Claude, ChatGPT and Gemini preferred to employ nuclear weapons rather than negotiate the end of crises.
The obvious problem is that this is today. With the investment made by governments and their struggle with big tech, the future is cloudy. "The only certainty we have is that we deal with uncertainty", summarizes the Swede Nilsson in a Socratic way.
