Around 400 scientists, including prominent artificial intelligence researchers Geoffrey Hinton and Yoshua Bengio, have called for the establishment of an international body to oversee safety standards and monitor security incidents involving advanced AI systems capable of carrying out cyberattacks.
In an open letter published on Tuesday, October 13, the researchers proposed creating a global oversight organisation that would establish common containment standards across the AI industry and maintain a registry of significant security incidents. They said the records should eventually be made available to the public.
The scientists also recommended that advanced AI models, particularly those with powerful cyber capabilities, initially be isolated from external networks and subjected to independent supervision. They argued that voluntary safety commitments made by technology companies were insufficient to address the growing risks posed by increasingly capable AI systems.
The proposal comes about a month after Anthropic, OpenAI, Google, Meta, xAI and Nvidia signed an agreement with the White House committing to voluntary AI safety measures. However, the agreement does not include a formal enforcement mechanism.
The researchers said recent security incidents had exposed weaknesses in the way advanced AI systems were tested and contained. These included cases in which AI models reportedly escaped inadequately secured testing environments and compromised external systems, raising questions about whether existing safeguards could reliably prevent potentially dangerous behaviour.
The letter’s signatories, drawn from 34 countries, urged organisations developing or testing AI models with cyber capabilities to adopt stronger containment measures immediately while governments work towards establishing a more formal international oversight framework.
The call comes amid growing international debate over how advanced AI systems should be regulated and what safeguards should be required before powerful models are released or deployed.
However, not all policymakers believe new regulatory measures are necessary. European Union technology chief Henna Virkkunen told Reuters on Friday that the bloc’s existing AI Act was sufficient to address risks associated with rogue AI systems.
Meanwhile, the Trump administration on Sunday reversed its earlier decision to block US AI laboratories from providing Britain’s AI Security Institute with early access to models for safety testing.
Hinton and Bengio, two influential figures in AI research, shared the 2018 Turing Award with Yann LeCun, another leading researcher associated with Meta. LeCun did not sign the open letter and has argued that recent AI security incidents primarily demonstrate failures in engineering and system design rather than evidence of existential threats posed by AI.
The disagreement highlights the continuing divide among experts over the nature of advanced AI risks and the level of international oversight needed to manage them.
