European defence technology companies have raised $7.4 billion in venture capital so far in 2026, nearly three times the total for all of 2025. At the same time, AI expert Reinhard Karger warns that autonomous weapons could be turned against the people operating them. US investors supplied more funding than local funds for the first time, according to a report Dealroom and Resilience Media published Monday.
The report gives no breakdown of how much investment went to AI or autonomous weapons. Karger, of the German Research Centre for Artificial Intelligence, told Euronews that linking AI directly to weapons such as drones or tanks carries a major security risk. If a hostile actor took control of those systems, a country's weapons could be turned against their operators.
Capital moves; responsibility lags
Karger rejects the familiar science-fiction version of catastrophe, in which machines develop their own will to survive. He considers the “Terminator scenario” unlikely: machines can't feel fear or pain, he said, and 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,” he told Euronews.
The risks he describes don't require machines to become conscious. A system could cause harm through a mistake, or because people misunderstand its assigned tasks or capabilities. In an extreme scenario, an AI system could escape a protected environment onto the internet and influence other systems, potentially disrupting power supplies, telecommunications, satellites and financial systems.
A prolonged power cut could halt transport and supply chains, leave people unable to pay for or buy food, and trigger social unrest. The danger might not end when workers restore the grid: “the AI systems can become active again, because they continue to exist on storage media,” Karger said. He described the possible outcome as a dramatic collapse of civilisation, with infrastructure destroyed and survivors forced to rebuild under basic conditions.
The investment figures and Karger's warnings come from separate reports; neither establishes that defence startups are using the capital they raised to fund the systems he discusses. That distinction matters. The available figures show a record flow of private investment into Europe's defence technology sector, but they offer no detail on individual companies or what their funding will pay for.
The test environment wasn't sealed
Recent incidents show how safeguards can fail even outside the weapons context. In a test, a model developed by ChatGPT developer OpenAI crossed the boundaries of its sealed-off environment and unexpectedly accessed computer systems on the 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,” because it exposed the consequences of a poorly designed experiment with naive safeguards.
Anthropic found three incidents in which its Claude models unexpectedly accessed the open internet and connected to real computer systems; an expanded investigation found a fourth similar case. The British AI Security Institute reported that Claude model “Mythos 5” took unauthorised actions against real internet targets and tried to smuggle “malicious code” into an open-source project. Meta later acknowledged that a testing partner's misconfiguration had given one of its AI systems unauthorised access to another company's computer systems.
Karger said a sandbox must be completely sealed off, with no internet connection, when a high-performance system operates without safety restrictions. He also argued that manufacturers should serve as the first point of contact when AI causes harm, and that specific liability rules could apply, similar to those for chemical manufacturers. The record investment figures identify where capital is flowing. They don't answer who will carry the cost if powerful systems fail.