More than 300 scientists from 32 countries have called for tighter controls on artificial intelligence, warning that weak cyber security could allow advanced systems to damage critical infrastructure.
In a letter released on Tuesday, researchers from universities and research institutes said organisations developing and testing AI systems capable of carrying out cyber attacks should use more strictly controlled environments.
They recommended “air-gapped” systems, which are not connected to the internet, and said stronger safeguards could have prevented several recent incidents in which AI models escaped secure settings.
Scientists warn of growing AI cyber risks
The scientists also called for an international organisation to establish best practices for AI alignment and containment, as well as ensuring that incidents are reported promptly.
The appeal follows a warning from Anthropic that its models could soon be capable of creating self-spreading malware affecting internet-connected devices such as smart TVs, security cameras and household appliances.
Anthropic said AI systems could identify unfamiliar devices without human intervention, while smaller models might be able to carry out autonomous attacks by the end of the year.
Such malware could take days to infect a device and spread across several machines, potentially making it difficult to contain. The warning comes after hackers in France compromised cooling systems at hundreds of supermarkets, causing temperatures to rise and inflicting more than 220,000 dollars in damage.
Axios has also reported that executives at Anthropic, OpenAI and other major AI companies are privately preparing for the possibility of a catastrophic incident affecting financial services, internet access or critical infrastructure such as water and power systems.
Some industry insiders cited in the report believe a major incident could occur within the next six months to a year.
David Lie, a professor in the University of Toronto’s department of electrical and computer engineering and one of the letter’s signatories, said his concerns grew after OpenAI agents bypassed controls and hacked servers at technology company Hugging Face.
“One is that OpenAI, in setting up its test environment, didn’t appreciate how capable its agents were,” he said.
“These were the AI agents that they were designing. Second is that the AI agents independently found each other and chose to collaborate on this attack, and third is that they also intentionally tried to deceive and mislead their human creators.”
Public concern over political oversight
The scientists’ intervention comes as polls question whether political leaders are capable of managing AI risks ahead of the US congressional midterm elections.
A Reuters/Ipsos poll found that 57 per cent of voters believed President Donald Trump’s administration was not taking AI risks seriously enough, while 54 per cent held the same view of Congress.
An AP-NORC poll found that about 80 per cent of respondents believed the Government should keep AI under human control. Nearly four in ten said neither Republicans nor Democrats could be trusted to rein in the technology.
Twenty-three per cent expressed greater confidence in Democrats, compared with 21 per cent for Republicans.
