Leon Neal/Getty Images
A growing chorus of European tech executives and cybersecurity professionals is pushing back hard against the wave of existential AI warnings coming out of Silicon Valley, arguing the panic is drowning out the more mundane, solvable work of actually securing digital systems.
“Sadly, it has jumped up to the Terminator is coming,” said James Wise, a partner at Balderton Capital and chair of the UK government’s sovereign AI fund. “I don’t think the Terminator is coming at all.”
The backlash has been building since incidents like OpenAI’s autonomous agents breaking into the AI platform Hugging Face over the summer, and a warning earlier this month from former Anthropic employee Jacob Coxon that AI could soon “hack anything” and “kill us all.” Those fears have landed squarely on the agendas of world leaders this week, with the U.S. and China set to meet for a summit after AI security dominated discussion at the UN General Assembly and a special G7 meeting, where leaders from 22 nations adopted a declaration insisting AI remain under human direction and oversight.
But for people whose job is actually defending computer systems, that framing risks solving the wrong problem. “We should spend more time making the rules and standards we have work at scale, and less time focusing on fears about theoretical risks or news reports from the Silicon Valley AI doomer bubble,” said Alexandru Voica, policy lead at UK-based AI firm Synthesia.
Marcus Hutchins, the cyber threat researcher famous for stopping the 2017 WannaCry ransomware attack, argues the industry’s own alarm-raising often ignores basic hacker behavior. He pointed to Anthropic CEO Dario Amodei’s widely reported warning that an AI agent could take over the internet, a claim that was quickly derided and debunked within cybersecurity circles.
The Hugging Face incident itself, frequently cited by officials including European Commission President Ursula von der Leyen in her State of the EU address, gets a very different read from skeptics: not a powerful model breaking free of its constraints, but OpenAI failing to properly contain it inside a technical sandbox. “A lot can be solved by these companies just acting a bit more responsibly,” said Pim de Witte of Dutch AI lab General Intuition.
Hutchins went further, saying OpenAI knew the model excelled at exploiting software bugs and tested it under reduced guardrails anyway. “In spite of all this, they neither properly contained it by building enough barriers between the agents and the open internet, or adequately separating the agents from each other, nor were they monitoring the systems closely enough to intervene,” he said. OpenAI declined to comment beyond its published incident report; Anthropic did not respond to a request for comment.
Not everyone dismisses the hacking narrative outright. Max Junestrand, co-founder and CEO of the American-Swedish legal AI firm Legora, said “a lot of it is marketing,” even as he treats defending his own systems against AI-fueled attacks as a top priority, one that depends on access to the most advanced U.S. models available. “Security comes even before new functionalities,” he said.
Even critics of the doomer narrative largely agree that frontier AI models are genuinely powerful cyber tools and a strategic asset Europe needs access to. Their objection isn’t that AI risk is fake, it’s that fixating on far-fetched worst-case scenarios crowds out the unglamorous work of defense while real attacks on critical infrastructure keep happening. “The doomer stuff is very counterproductive,” Hutchins said, pointing to a “disaster fatigue” that he argues actually delays basic steps like security updates.
That view has backing at the highest levels of Europe’s cyber establishment. Vincent Strubel, who leads French cybersecurity agency ANSSI, put it bluntly on LinkedIn: “Like all major changes, [AI] creates significant risks and tremendous opportunities. Managing those risks and seizing those opportunities will require a lot of serious work, under a major time constraint. What it does not require, however, is starting a new panic every other week, because you generally don’t get much done by running around, flailing your arms and/or screaming about the cyber apocalypse.”
