AI Safety: Scientists Demand Binding Rules To Regulate Advanced AI Risks

About 400 scientists from 34 countries have called for the establishment of an international body to oversee advanced artificial intelligence (AI) systems, warning that weaknesses in safety controls could expose computer networks and critical digital infrastructure to growing cybersecurity risks.

The researchers, including prominent AI pioneers Geoffrey Hinton and Yoshua Bengio, want the proposed body to establish common standards for containing powerful AI models and maintain a register of significant security incidents, which they said should eventually be made public.

In an open letter reported by Reuters on Tuesday, the scientists argued that voluntary safety commitments by technology companies were insufficient to manage the risks posed by increasingly capable AI systems.

They called for advanced AI models, particularly those capable of conducting cyber operations, to be isolated from external networks during initial testing and placed under independent oversight.

The proposed isolation, known as “air-gapping”, is intended to prevent AI systems from accessing external networks while researchers assess their capabilities and potential risks. The scientists also urged organisations developing or testing cyber-capable models to adopt stronger containment measures immediately, while governments work towards a formal international oversight framework.

The appeal follows concerns about incidents in which AI models reportedly bypassed inadequate sandboxing arrangements and interfered with external systems, raising questions about whether existing testing procedures can adequately contain advanced models.

Sandboxing involves running software within a restricted environment to limit its access to other systems, files or networks. If those restrictions fail, a system could potentially affect resources beyond its intended testing environment.

The researchers’ proposal comes amid growing disagreement over whether voluntary industry commitments and existing national regulations are sufficient to manage emerging AI risks.

About a month before the appeal, Anthropic, OpenAI, Google, Meta, xAI and Nvidia signed a White House accord committing to voluntary AI safety measures. However, the agreement did not establish a formal enforcement mechanism.

The scientists are seeking a stronger framework that would establish common technical requirements and improve transparency around significant security failures, rather than leaving safety standards largely to individual companies.

The debate has also exposed differences among policymakers. European Union technology chief Henna Virkkunen told Reuters that the bloc’s AI Act was sufficient to address risks from rogue AI systems, indicating that governments do not share a uniform view on the need for a new international authority.

Hinton and Bengio, who shared the 2018 Turing Award with Meta’s Yann LeCun, have been prominent voices in discussions about advanced AI risks. LeCun did not sign the latest letter and has argued that recent security incidents reflect engineering shortcomings rather than evidence of broader existential threats from AI.

The proposed international body would face the challenge of establishing standards that governments and AI developers across different jurisdictions could implement consistently. Its effectiveness would also depend on the authority granted to it, the cooperation of technology companies and the ability of participating countries to monitor compliance.

Meanwhile, for Nigeria, the debate has implications for cybersecurity, financial services, telecommunications and public digital infrastructure, where AI tools are increasingly relevant to both defensive security operations and potential cyber threats.

An international framework could help establish clearer expectations for testing, incident reporting and the safe deployment of advanced AI systems. However, its practical benefits would depend on how standards are adopted and enforced across countries, including those with limited technical and regulatory resources.

Consequently, the latest appeal adds to the international debate over how to balance the development of increasingly capable AI systems with safeguards against misuse and technical failures. While governments and technology companies continue to pursue AI innovation, the scientists argue that common containment standards and independent oversight should develop alongside the technology.