Scientists urge tighter AI controls to prevent global cyber risks

What the report says

More than 300 scientists from 32 countries have signed an open letter urging stricter safety and cybersecurity controls for advanced artificial intelligence systems, according to Al Jazeera. The appeal, published Tuesday, says AI developers should use more tightly restricted testing environments and stronger containment measures to reduce the chance that models could behave unpredictably or be used for cyberattacks.

The researchers warned that weak security practices could expose critical infrastructure to damage if AI systems are able to escape controlled settings. They recommended that companies developing and evaluating powerful models, including firms such as OpenAI and Anthropic, keep sensitive testing work in environments that are not connected to the internet and adopt clearer incident reporting and alignment safeguards.

Al Jazeera reported the letter amid a wider debate over AI risk, including fresh warnings from Anthropic that future models could help create self-spreading malware targeting internet-connected devices. The article also cited broader concern among the public and political leaders in the United States, with recent polling showing many voters doubt current officials can manage AI risks effectively ahead of the midterm elections.

The scientists also called for the creation of an international body to help define best practices for AI containment and safety. The issue matters because experts say a serious AI-related cyber incident could affect everyday devices, financial services, and potentially essential infrastructure such as power or water systems.