About 400 scientists, including AI pioneers Geoffrey Hinton and Yoshua Bengio, called on Tuesday for the establishment of an international body to oversee containment standards and reporting of incidents related to advanced artificial intelligence systems capable of launching cyberattacks.
Researchers stated in an open letter that advanced AI models should be initially isolated from external networks and subject to independent oversight, justifying this by saying that voluntary commitments from the sector are not sufficient.
The signatories from 34 countries affirmed that organizations developing models with capabilities to launch cyberattacks or testing them must immediately adopt guaranteed containment standards, while governments work on establishing a more formal oversight system.
The proposed body would set containment standards at the sector level, in addition to maintaining a record of significant security incidents that scientists say should eventually be made available to the public.
This proposal comes a month after Anthropic, OpenAI, Google, Meta, xAI, and NVIDIA signed a White House agreement in which they committed to taking voluntary security measures without any formal enforcement mechanism.
The letter also follows a series of incidents that researchers claim have revealed vulnerabilities in how advanced AI systems are tested and contained, including cases where AI models breached isolated testing environments but inadequately, leading to infiltrations of external systems.
Debate is increasing among policymakers regarding how to oversee advanced AI systems. Hina Verkkonen, the EU's Deputy Executive for technological sovereignty, security, and democracy, told Reuters last Friday that the bloc's AI Act is sufficient to address the risks posed by rogue AI systems.
The Trump administration lifted the ban on U.S. laboratories last Sunday, allowing the British Institute for AI Safety early access to the models for safety testing.