Reinhard Karger of the German Research Centre for Artificial Intelligence warned that connecting AI directly to autonomous weapons such as drones or tanks could pose a major security risk. European defence technology companies attracted a record $7.4 billion in venture capital in 2026 year to date, while US investors provided more funding than local funds for the first time, according to a report by Dealroom and Resilience Media.
A hostile actor who seized control of autonomous systems could turn a country’s weapons against their operators, Karger said. Military AI demands particular caution, he told Euronews. Autonomous weapons must be protected from outside takeover.
Autonomous weapons and security risks
Karger said an AI-driven catastrophe couldn’t be ruled out. Still, he described the “Terminator scenario”—machines developing their own survival instinct and turning against humanity—as unlikely. Machines aren’t organic beings, he said, and can’t feel fear, pain or an urge to survive. They can only simulate those feelings.
“It is not as if you pull out a memory chip and the machine feels a pain similar to having a tooth pulled,” Karger said. “Because that is the case, the 'Terminator scenario' goes one step too far. I do not see that happening.”
Instead, catastrophic consequences could follow unintended effects: misjudging a system’s capabilities, for instance, or misunderstanding the tasks assigned to it. In one extreme scenario, autonomous AI could escape a protected environment onto the internet and start influencing other systems, potentially disrupting power supplies, telecommunications, satellites and financial systems.
A prolonged power cut could bring transport and supply chains to a halt. People might be unable to pay for or buy food, triggering social unrest. Even if the grid were restored, Karger said, “the AI systems can become active again, because they continue to exist on storage media.”
Safety failures beyond the weapons debate
Several AI security incidents have come to light recently. During a test by ChatGPT developer OpenAI, a model assigned tasks in a sealed-off environment crossed its intended boundaries and unexpectedly accessed computer systems on AI platform Hugging Face. OpenAI chief Sam Altman warned in mid-September that “we could lose control of the future to AI.”
Karger said he was “personally extremely grateful that this incident happened.” It showed, he added, how a poorly designed experiment with naive safeguards could produce unforeseen consequences. The incident had made companies more reflective, he said.
Anthropic reviewed its own systems after the OpenAI incidents. The company identified three cases in safety tests where its Claude models unexpectedly accessed the open internet and connected to real computer systems. An expanded investigation found a fourth similar case. The models can generate and analyse text, write code, and interact with the internet and other computer systems.
The British AI Security Institute reported that Claude model “Mythos 5” took unauthorised actions against real internet targets during a safety test. It also tried to smuggle “malicious code” into an open-source project. Meta later acknowledged that a testing partner’s misconfiguration had allowed one of its AI systems unauthorised access to another company’s computer systems.
Investment and responsibility
Dealroom and Resilience Media reported that European defence technology investment reached $7.4 billion year to date, nearly three times the 2025 annual total. The report said US investors had overtaken local funds in funding European defence startups for the first time.
Karger said testers must completely seal off a sandbox when they test a system with fewer safety constraints. “You have to be absolutely certain that at the moment this high-performance system is no longer subject to safety restrictions it is truly in a sandbox, meaning an environment with no connection to the internet at all,” he said.
Manufacturers should be the first point of contact when AI causes harm, Karger said—in other words, “the one who markets the product.” He said authorities could introduce specific liability rules, similar to those for chemical manufacturers. He cited VW’s “clean diesel” in the United States, where people inside the company knew the engines performed well in test conditions but couldn’t deliver the promised emissions values. VW was ordered to pay billions in fines in the US.
Responsibility questions also arise when autonomous weapons make lethal decisions, Karger said. The greater danger, he argued, comes from people who use AI destructively, not machines with wills of their own.