AI-controlled weapons are here and scary - but there is a solution

No time? blue News summarizes for you

• The war in Ukraine has revolutionized AI warfare.
• The fear that AI could wipe out humanity has increased during this time.
• In simulated war games, AI relies on nuclear strikes 95 percent of the time. However, expert Katrin J. Yuan believes: "We are not there yet."
• Johannes Lederer from the University of Hamburg has a pragmatic suggestion.

Summary created with

The Russian invasion of Ukraine has revolutionized drone warfare. Since 2024, 75 to 95 percent of damage on the ground has been caused by remote-controlled flying objects, according to Polytechnique Inside: Kiev is expected to have produced around five million drones by 2025.

But these drones also need pilots to control them - and electronic jamming often prevents them from finding their target. Artificial intelligence (AI) is intended to remedy these problems: The US company Shield AI has just announced a corresponding cooperation with Kiev.

Shield AI has developed a system called Hivemind, or swarm mind. It is to be implemented in Ukrainian drones and enable them to independently change routes and coordinate themselves in the event of danger. A swarm of drones controlled in this way should be able to divide up targets among themselves and engage them independently despite jamming.

The practical benefits in Ukraine

Humans only specify the target: The AI decides how this should be achieved. The American-Ukrainian company Swarmer is also working on such a solution. The reason: in an environment such as the embattled city of Pokovsk in Donetsk, "most Western systems don't work" because Russia interferes with them, writes Forbes.

There are also innovations on the ground: Using robots reduces the risk of losing people. The Ukrainian armed forces are therefore increasingly relying on the use ofunmanned ground vehicles (UGVs). "Modern UGVs are semi-autonomous," explains a soldier called Afghan to the BBC.

He continues: "They can move on their own, they can observe and detect the enemy. But the decision to open fire is still made by a human." The aim is to avoid fatal mistakes because the systems are not yet flawless.

At sea, AI-controlled naval drones are capable of attacking Russian ships in the Black Sea and beyond, even without radio contact. The concept is already being copied.

What works in Ukraine is being noticed in the West: The French military has just signed a contract with the Paris-based company Mistral to establish its AI in the armed forces. However, this is less about the concept of the kill chain, which describes the structure of an attack, and more about data processing.

Fear of being wiped out by AI

The Mistral AI is intended to improve the flow of information, develop tools for processing documents, keep logs, overlook logistics and translate texts as well as read and process data, writes "Army Recognition". At the same time, Paris is relying on a national solution to preserve France's independence.

But even an AI that only deals with data can be deadly. This is demonstrated by the US attack on a girls' school in Minab, Iran. Palantir Technologies ' Maven AI is responsible for the choice of target: US billionaire Peter Thiel co-founded the company.

So are the doomsday scenarios depicted in films such as "Metropolis", "The Matrix" and "Terminator" true after all? Some experts warn that intelligent machines could really do away with humanity. Geoffrey Hinton, the "godfather of AI", tells CNBC: "I often say there's a 10 to 20 percent chance that AI will wipe us out."

Daniel Kokotajlo from the US NGO AI Futures Project goes even further. "We think there's a 70 percent chance that all humans will die," he tells the Daily Show.

Kokotajlo used to work for OpenAI, the parent company of ChatGPT - and left the company in protest at the corporate culture. He believes that in just five years' time, an AI could exterminate its creators with biological weapons, for example.

"We're not there yet"

A study by King's College London, which was published at the end of February, examined how AI acts in war games: "Nuclear escalation was almost ubiquitous," summarizes Professor Kenneth Payne. "Tactical nuclear weapons were used in 95 percent of the games and 76 percent achieved strategic nuclear threats."

The Claude and Gemini models treated nuclear weapons as legitimate instruments. GPT-5.2 stood out somewhat because the AI was limited to military targets and avoided attacks on population centers.

"I've been advocating responsible AI for years," Katrin J. Yuan told blue News, commenting on the result and saying: "We're not there yet." AI imitates humans, but thinks "completely differently", explains the expert from the Swiss Future Instituteand Chair of the AI Future Council.

She cites programming to maximize human happiness as an example: AI could decide that we humans are standing in the way of our own happiness - with corresponding consequences.

Question of input and moral guidelines

The input of AI is of course the be-all and end-all: the basis on which it stands must be right. On the one hand, this applies to simple data: according to the Guardian, the problem with the attack on the girls' school in Minab was that the Maven AI had not been fed with updated information.

Expert explains

Everything you need to know about AI in weapons

On the other hand, more general guard rails are also necessary. Sometimes the manufacturers even take care of this themselves: Anthropic and OpenAI, the companies behind Claude and ChatGPT, are looking for experts in biological and chemical warfare agents and explosives in job advertisements. As the BBC reports, their job is to ensure that the AI cannot be misused by terrorists to build dirty bombs.

Dispute with the Pentagon

AI for killing and surveillance - Anthropic sues US government

Anthrophic has also clashed with the Pentagon because it did not want to give the armed forces full access to the Claude code - allegedly out of concern that the program would be used for illegal activities. A legal dispute between the parties is ongoing.

Expert brings international authority into play

The problem: if individual players in individual countries limit themselves and others do not, there is still a risk to people. "It's an international question, of course," emphasizes Johannes Lederer, who studied mathematics at ETH Zurich and now teaches in Hamburg, in an interview with blue News.

A "Terminator" figure made of scrap metal at a checkpoint in Kiev, photographed on March 25, 2022 shortly after the outbreak of war.

KEYSTONE

The data scientist admits that different countries have different values, but also points to an example that could set a precedent: the International Atomic Energy Agency (IAEA) in Vienna.

It has managed "relatively well" to limit the proliferation of nuclear weapons: "I can imagine the same with AI in weapons," he says in the video at the top.