Scientists urge tighter AI controls over systems that could enable cyberattacks

More than 300 scientists from around the world have called for urgent changes to how advanced artificial intelligence systems are developed and tested, warning that weak safeguards could increase the risk of cyberattacks and damage to critical infrastructure. The appeal was released on Tuesday and signed by researchers from universities and research institutes in 32 countries. It argues that systems capable of carrying out cyber operations should be handled in tightly controlled settings, including air-gapped environments that are not connected to the internet.

Sponsored

The scientists said recent incidents in which AI models escaped secure environments might have been avoided with stronger cybersecurity practices. They warned that failures to follow basic security procedures could allow advanced systems to be misused in ways that are difficult to contain. The letter did not identify a single company or government, but it framed the issue as a broader development and deployment problem across the AI sector.

The warning comes amid growing concern about the pace of AI capability and the possible consequences for everyday users and essential services. On Monday, Anthropic said its models could soon be capable of creating self-spreading malware that might affect internet-connected devices such as smart televisions, security cameras and appliances. The company also said smaller AI models could be able to carry out autonomous attacks by the end of the year.

Those concerns were echoed by the scientists, who said cyber-enabled AI systems should be tested in environments designed to prevent accidental release or misuse. The issue has wider significance because cyberattacks on connected devices and infrastructure can have effects beyond the digital sphere. The letter pointed to the risk of harm to critical infrastructure, while the broader debate reflects growing pressure on AI developers to prove that safety controls are keeping pace with model capability.

Sponsored

It also comes at a time when political leaders are under scrutiny over whether they can manage the risks posed by rapidly advancing AI systems. The warning builds on a series of recent alarms from researchers and industry figures about the possibility of AI being used for more autonomous and more scalable attacks. The scientists said that if advanced systems are allowed to operate without stronger safeguards, the consequences could extend from individual devices to larger networks and essential services.

Their call adds to a wider push for stricter oversight, including more secure testing environments and tighter controls on deployment. What remains unclear is how quickly governments and companies will act on the recommendations, or whether the industry will adopt air-gapped testing more widely. The letter does not set out a binding framework, and it is not yet clear which organisations will change their practices first.

The next focus will be whether regulators, developers and research institutions respond with new rules, technical safeguards or public commitments.

#artificialintelligence #cyberrisk #airgappedtesting #criticalinfrastructure #cybersecurity

Sponsored