BRUSSELS — Amid fears that AI models could destroy humanity, top European tech leaders and cybersecurity experts have an increasingly loud response: Ignore the hype, and get on with the hard work of keeping the digital world safe. That prosaic call is at the heart of an increasingly bitter backlash in Europe’s tech industry against the U.S.-generated hype about AI extinction. All the “Chicken Little” warnings, say both executives and cyber professionals, are getting in the way of the hard work of reinforcing defenses against the new AI reality.
The U.S. companies building the most advanced AI models have largely been responsible for raising the specter that artificial intelligence could lead to the demise of the human race. Fears have been peaking since incidents like OpenAI’s autonomous agents breaking into AI platform Hugging Face this summer, and a warning earlier this month by former Anthropic employee Jacob Coxon that AI could soon “hack anything” and “kill us all.” “Sadly, it has jumped up to the Terminator is coming,” James Wise, a partner at Balderton Capital and chair of the U.K. government’s sovereign AI fund, said of the current narrative. President Donald Trump and Chinese leader Xi Jinping meeting for a Thursday summit after AI security was debated at the United Nations General Assembly and a special G7 meeting.
Leaders from 22 nations adopted a declaration on Monday saying that AI must remain under human direction, oversight and control, and suggesting the creation of a global supervisory regime. Yet for those tasked with defending computer systems, the daunting chorus of warnings about advanced AI models — and the political scramble to respond — risks applying the wrong solution to the wrong problem. A recent example of AI voices arguably missing the mark is a warning by Anthropic CEO Dario Amodei that an AI agent could take over the internet.
That claim, which was widely reported, was derided and debunked in cybersecurity circles. Skeptics also point to the constant references to the Hugging Face incident — which even European Commission President Ursula von der Leyen cited in her State of the EU address — as emblematic of the hysteria regarding would-be runaway AI models. Rather than depicting a powerful AI model breaking free of its constraints, the assault on Hugging Face arguably offers a much more mundane lesson: That OpenAI failed to properly contain the model in a technical environment known as a “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, which develops models that are aware of their surroundings.
Hutchins said OpenAI knew the model was very good at exploiting software bugs, and yet challenged it to do just that in a test conducted with reduced guardrails. OpenAI declined to comment, pointing to its report on the incident. Anthropic did not respond to a request for comment.
Max Junestrand, co-founder and CEO of Legora, an American-Swedish firm that develops collaborative AI tools for lawyers, said of the hacking incidents: “A lot of it is marketing.” Still, securing Legora’s operations against AI-fueled hacking attacks is a top priority for Junestrand, who emphasized that having access to the latest and most advanced U.S. models is key to that endeavor: “Security comes even before new functionalities.” Indeed, even critics and skeptics of the doomer narrative agree that frontier AI models are uniquely powerful cyber tools, as well as a strategic asset that Europe needs to be able to access. Moreover, few deny that the current AI capacities being developed are on a whole new level, with their potential impacts — both positive and negative — hard to predict. Yet critics of the current narrative say that exaggerating AI capabilities or the extent to which hackers are using them — and the parlor game of guessing theoretical outcomes — is unhelpful when real-world cyberattacks are already disrupting critical infrastructure.
Extract — continue reading at the source.
